Pull motherboard into release branch
[linux-2.6] / arch / i386 / kernel / setup.c
1 /*
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
26 #include <linux/config.h>
27 #include <linux/sched.h>
28 #include <linux/mm.h>
29 #include <linux/mmzone.h>
30 #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/platform_device.h>
38 #include <linux/console.h>
39 #include <linux/mca.h>
40 #include <linux/root_dev.h>
41 #include <linux/highmem.h>
42 #include <linux/module.h>
43 #include <linux/efi.h>
44 #include <linux/init.h>
45 #include <linux/edd.h>
46 #include <linux/nodemask.h>
47 #include <linux/kexec.h>
48 #include <linux/crash_dump.h>
49 #include <linux/dmi.h>
50 #include <linux/pfn.h>
51
52 #include <video/edid.h>
53
54 #include <asm/apic.h>
55 #include <asm/e820.h>
56 #include <asm/mpspec.h>
57 #include <asm/setup.h>
58 #include <asm/arch_hooks.h>
59 #include <asm/sections.h>
60 #include <asm/io_apic.h>
61 #include <asm/ist.h>
62 #include <asm/io.h>
63 #include "setup_arch_pre.h"
64 #include <bios_ebda.h>
65
66 /* Forward Declaration. */
67 void __init find_max_pfn(void);
68
69 /* This value is set up by the early boot code to point to the value
70    immediately after the boot time page tables.  It contains a *physical*
71    address, and must not be in the .bss segment! */
72 unsigned long init_pg_tables_end __initdata = ~0UL;
73
74 int disable_pse __devinitdata = 0;
75
76 /*
77  * Machine setup..
78  */
79
80 #ifdef CONFIG_EFI
81 int efi_enabled = 0;
82 EXPORT_SYMBOL(efi_enabled);
83 #endif
84
85 /* cpu data as detected by the assembly code in head.S */
86 struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
87 /* common cpu data for all cpus */
88 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
89 EXPORT_SYMBOL(boot_cpu_data);
90
91 unsigned long mmu_cr4_features;
92
93 #ifdef  CONFIG_ACPI
94         int acpi_disabled = 0;
95 #else
96         int acpi_disabled = 1;
97 #endif
98 EXPORT_SYMBOL(acpi_disabled);
99
100 #ifdef  CONFIG_ACPI
101 int __initdata acpi_force = 0;
102 extern acpi_interrupt_flags     acpi_sci_flags;
103 #endif
104
105 /* for MCA, but anyone else can use it if they want */
106 unsigned int machine_id;
107 #ifdef CONFIG_MCA
108 EXPORT_SYMBOL(machine_id);
109 #endif
110 unsigned int machine_submodel_id;
111 unsigned int BIOS_revision;
112 unsigned int mca_pentium_flag;
113
114 /* For PCI or other memory-mapped resources */
115 unsigned long pci_mem_start = 0x10000000;
116 #ifdef CONFIG_PCI
117 EXPORT_SYMBOL(pci_mem_start);
118 #endif
119
120 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
121 int bootloader_type;
122
123 /* user-defined highmem size */
124 static unsigned int highmem_pages = -1;
125
126 /*
127  * Setup options
128  */
129 struct drive_info_struct { char dummy[32]; } drive_info;
130 #if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_HD) || \
131     defined(CONFIG_BLK_DEV_IDE_MODULE) || defined(CONFIG_BLK_DEV_HD_MODULE)
132 EXPORT_SYMBOL(drive_info);
133 #endif
134 struct screen_info screen_info;
135 EXPORT_SYMBOL(screen_info);
136 struct apm_info apm_info;
137 EXPORT_SYMBOL(apm_info);
138 struct sys_desc_table_struct {
139         unsigned short length;
140         unsigned char table[0];
141 };
142 struct edid_info edid_info;
143 EXPORT_SYMBOL_GPL(edid_info);
144 struct ist_info ist_info;
145 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
146         defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
147 EXPORT_SYMBOL(ist_info);
148 #endif
149 struct e820map e820;
150
151 extern void early_cpu_init(void);
152 extern void generic_apic_probe(char *);
153 extern int root_mountflags;
154
155 unsigned long saved_videomode;
156
157 #define RAMDISK_IMAGE_START_MASK        0x07FF
158 #define RAMDISK_PROMPT_FLAG             0x8000
159 #define RAMDISK_LOAD_FLAG               0x4000  
160
161 static char command_line[COMMAND_LINE_SIZE];
162
163 unsigned char __initdata boot_params[PARAM_SIZE];
164
165 static struct resource data_resource = {
166         .name   = "Kernel data",
167         .start  = 0,
168         .end    = 0,
169         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
170 };
171
172 static struct resource code_resource = {
173         .name   = "Kernel code",
174         .start  = 0,
175         .end    = 0,
176         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
177 };
178
179 static 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
186 static 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
193 static 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
228 static 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
235 static struct resource video_ram_resource = {
236         .name   = "Video RAM area",
237         .start  = 0xa0000,
238         .end    = 0xbffff,
239         .flags  = IORESOURCE_BUSY | IORESOURCE_MEM
240 };
241
242 static 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
294 static 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
303 static 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
368 static void __init limit_regions(unsigned long long size)
369 {
370         unsigned long long current_addr = 0;
371         int i;
372
373         if (efi_enabled) {
374                 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) {
382                                 if (current_addr >= size) {
383                                         md->num_pages -=
384                                                 (((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                 current_addr = e820.map[i].addr + e820.map[i].size;
393                 if (current_addr < size)
394                         continue;
395
396                 if (e820.map[i].type != E820_RAM)
397                         continue;
398
399                 if (e820.map[i].addr >= size) {
400                         /*
401                          * This region starts past the end of the
402                          * requested size, skip it completely.
403                          */
404                         e820.nr_map = i;
405                 } else {
406                         e820.nr_map = i + 1;
407                         e820.map[i].size -= current_addr - size;
408                 }
409                 return;
410         }
411 }
412
413 static void __init add_memory_region(unsigned long long start,
414                                   unsigned long long size, int type)
415 {
416         int x;
417
418         if (!efi_enabled) {
419                 x = e820.nr_map;
420
421                 if (x == E820MAX) {
422                     printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
423                     return;
424                 }
425
426                 e820.map[x].addr = start;
427                 e820.map[x].size = size;
428                 e820.map[x].type = type;
429                 e820.nr_map++;
430         }
431 } /* add_memory_region */
432
433 #define E820_DEBUG      1
434
435 static void __init print_memory_map(char *who)
436 {
437         int i;
438
439         for (i = 0; i < e820.nr_map; i++) {
440                 printk(" %s: %016Lx - %016Lx ", who,
441                         e820.map[i].addr,
442                         e820.map[i].addr + e820.map[i].size);
443                 switch (e820.map[i].type) {
444                 case E820_RAM:  printk("(usable)\n");
445                                 break;
446                 case E820_RESERVED:
447                                 printk("(reserved)\n");
448                                 break;
449                 case E820_ACPI:
450                                 printk("(ACPI data)\n");
451                                 break;
452                 case E820_NVS:
453                                 printk("(ACPI NVS)\n");
454                                 break;
455                 default:        printk("type %lu\n", e820.map[i].type);
456                                 break;
457                 }
458         }
459 }
460
461 /*
462  * Sanitize the BIOS e820 map.
463  *
464  * Some e820 responses include overlapping entries.  The following 
465  * replaces the original e820 map with a new one, removing overlaps.
466  *
467  */
468 struct change_member {
469         struct e820entry *pbios; /* pointer to original bios entry */
470         unsigned long long addr; /* address for this change point */
471 };
472 static struct change_member change_point_list[2*E820MAX] __initdata;
473 static struct change_member *change_point[2*E820MAX] __initdata;
474 static struct e820entry *overlap_list[E820MAX] __initdata;
475 static struct e820entry new_bios[E820MAX] __initdata;
476
477 static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
478 {
479         struct change_member *change_tmp;
480         unsigned long current_type, last_type;
481         unsigned long long last_addr;
482         int chgidx, still_changing;
483         int overlap_entries;
484         int new_bios_entry;
485         int old_nr, new_nr, chg_nr;
486         int i;
487
488         /*
489                 Visually we're performing the following (1,2,3,4 = memory types)...
490
491                 Sample memory map (w/overlaps):
492                    ____22__________________
493                    ______________________4_
494                    ____1111________________
495                    _44_____________________
496                    11111111________________
497                    ____________________33__
498                    ___________44___________
499                    __________33333_________
500                    ______________22________
501                    ___________________2222_
502                    _________111111111______
503                    _____________________11_
504                    _________________4______
505
506                 Sanitized equivalent (no overlap):
507                    1_______________________
508                    _44_____________________
509                    ___1____________________
510                    ____22__________________
511                    ______11________________
512                    _________1______________
513                    __________3_____________
514                    ___________44___________
515                    _____________33_________
516                    _______________2________
517                    ________________1_______
518                    _________________4______
519                    ___________________2____
520                    ____________________33__
521                    ______________________4_
522         */
523
524         /* if there's only one memory region, don't bother */
525         if (*pnr_map < 2)
526                 return -1;
527
528         old_nr = *pnr_map;
529
530         /* bail out if we find any unreasonable addresses in bios map */
531         for (i=0; i<old_nr; i++)
532                 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
533                         return -1;
534
535         /* create pointers for initial change-point information (for sorting) */
536         for (i=0; i < 2*old_nr; i++)
537                 change_point[i] = &change_point_list[i];
538
539         /* record all known change-points (starting and ending addresses),
540            omitting those that are for empty memory regions */
541         chgidx = 0;
542         for (i=0; i < old_nr; i++)      {
543                 if (biosmap[i].size != 0) {
544                         change_point[chgidx]->addr = biosmap[i].addr;
545                         change_point[chgidx++]->pbios = &biosmap[i];
546                         change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
547                         change_point[chgidx++]->pbios = &biosmap[i];
548                 }
549         }
550         chg_nr = chgidx;        /* true number of change-points */
551
552         /* sort change-point list by memory addresses (low -> high) */
553         still_changing = 1;
554         while (still_changing)  {
555                 still_changing = 0;
556                 for (i=1; i < chg_nr; i++)  {
557                         /* if <current_addr> > <last_addr>, swap */
558                         /* or, if current=<start_addr> & last=<end_addr>, swap */
559                         if ((change_point[i]->addr < change_point[i-1]->addr) ||
560                                 ((change_point[i]->addr == change_point[i-1]->addr) &&
561                                  (change_point[i]->addr == change_point[i]->pbios->addr) &&
562                                  (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
563                            )
564                         {
565                                 change_tmp = change_point[i];
566                                 change_point[i] = change_point[i-1];
567                                 change_point[i-1] = change_tmp;
568                                 still_changing=1;
569                         }
570                 }
571         }
572
573         /* create a new bios memory map, removing overlaps */
574         overlap_entries=0;       /* number of entries in the overlap table */
575         new_bios_entry=0;        /* index for creating new bios map entries */
576         last_type = 0;           /* start with undefined memory type */
577         last_addr = 0;           /* start with 0 as last starting address */
578         /* loop through change-points, determining affect on the new bios map */
579         for (chgidx=0; chgidx < chg_nr; chgidx++)
580         {
581                 /* keep track of all overlapping bios entries */
582                 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
583                 {
584                         /* add map entry to overlap list (> 1 entry implies an overlap) */
585                         overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
586                 }
587                 else
588                 {
589                         /* remove entry from list (order independent, so swap with last) */
590                         for (i=0; i<overlap_entries; i++)
591                         {
592                                 if (overlap_list[i] == change_point[chgidx]->pbios)
593                                         overlap_list[i] = overlap_list[overlap_entries-1];
594                         }
595                         overlap_entries--;
596                 }
597                 /* if there are overlapping entries, decide which "type" to use */
598                 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
599                 current_type = 0;
600                 for (i=0; i<overlap_entries; i++)
601                         if (overlap_list[i]->type > current_type)
602                                 current_type = overlap_list[i]->type;
603                 /* continue building up new bios map based on this information */
604                 if (current_type != last_type)  {
605                         if (last_type != 0)      {
606                                 new_bios[new_bios_entry].size =
607                                         change_point[chgidx]->addr - last_addr;
608                                 /* move forward only if the new size was non-zero */
609                                 if (new_bios[new_bios_entry].size != 0)
610                                         if (++new_bios_entry >= E820MAX)
611                                                 break;  /* no more space left for new bios entries */
612                         }
613                         if (current_type != 0)  {
614                                 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
615                                 new_bios[new_bios_entry].type = current_type;
616                                 last_addr=change_point[chgidx]->addr;
617                         }
618                         last_type = current_type;
619                 }
620         }
621         new_nr = new_bios_entry;   /* retain count for new bios entries */
622
623         /* copy new bios mapping into original location */
624         memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
625         *pnr_map = new_nr;
626
627         return 0;
628 }
629
630 /*
631  * Copy the BIOS e820 map into a safe place.
632  *
633  * Sanity-check it while we're at it..
634  *
635  * If we're lucky and live on a modern system, the setup code
636  * will have given us a memory map that we can use to properly
637  * set up memory.  If we aren't, we'll fake a memory map.
638  *
639  * We check to see that the memory map contains at least 2 elements
640  * before we'll use it, because the detection code in setup.S may
641  * not be perfect and most every PC known to man has two memory
642  * regions: one from 0 to 640k, and one from 1mb up.  (The IBM
643  * thinkpad 560x, for example, does not cooperate with the memory
644  * detection code.)
645  */
646 static int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
647 {
648         /* Only one memory region (or negative)? Ignore it */
649         if (nr_map < 2)
650                 return -1;
651
652         do {
653                 unsigned long long start = biosmap->addr;
654                 unsigned long long size = biosmap->size;
655                 unsigned long long end = start + size;
656                 unsigned long type = biosmap->type;
657
658                 /* Overflow in 64 bits? Ignore the memory map. */
659                 if (start > end)
660                         return -1;
661
662                 /*
663                  * Some BIOSes claim RAM in the 640k - 1M region.
664                  * Not right. Fix it up.
665                  */
666                 if (type == E820_RAM) {
667                         if (start < 0x100000ULL && end > 0xA0000ULL) {
668                                 if (start < 0xA0000ULL)
669                                         add_memory_region(start, 0xA0000ULL-start, type);
670                                 if (end <= 0x100000ULL)
671                                         continue;
672                                 start = 0x100000ULL;
673                                 size = end - start;
674                         }
675                 }
676                 add_memory_region(start, size, type);
677         } while (biosmap++,--nr_map);
678         return 0;
679 }
680
681 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
682 struct edd edd;
683 #ifdef CONFIG_EDD_MODULE
684 EXPORT_SYMBOL(edd);
685 #endif
686 /**
687  * copy_edd() - Copy the BIOS EDD information
688  *              from boot_params into a safe place.
689  *
690  */
691 static inline void copy_edd(void)
692 {
693      memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
694      memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
695      edd.mbr_signature_nr = EDD_MBR_SIG_NR;
696      edd.edd_info_nr = EDD_NR;
697 }
698 #else
699 static inline void copy_edd(void)
700 {
701 }
702 #endif
703
704 /*
705  * Do NOT EVER look at the BIOS memory size location.
706  * It does not work on many machines.
707  */
708 #define LOWMEMSIZE()    (0x9f000)
709
710 static void __init parse_cmdline_early (char ** cmdline_p)
711 {
712         char c = ' ', *to = command_line, *from = saved_command_line;
713         int len = 0;
714         int userdef = 0;
715
716         /* Save unparsed command line copy for /proc/cmdline */
717         saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
718
719         for (;;) {
720                 if (c != ' ')
721                         goto next_char;
722                 /*
723                  * "mem=nopentium" disables the 4MB page tables.
724                  * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM
725                  * to <mem>, overriding the bios size.
726                  * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from
727                  * <start> to <start>+<mem>, overriding the bios size.
728                  *
729                  * HPA tells me bootloaders need to parse mem=, so no new
730                  * option should be mem=  [also see Documentation/i386/boot.txt]
731                  */
732                 if (!memcmp(from, "mem=", 4)) {
733                         if (to != command_line)
734                                 to--;
735                         if (!memcmp(from+4, "nopentium", 9)) {
736                                 from += 9+4;
737                                 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
738                                 disable_pse = 1;
739                         } else {
740                                 /* If the user specifies memory size, we
741                                  * limit the BIOS-provided memory map to
742                                  * that size. exactmap can be used to specify
743                                  * the exact map. mem=number can be used to
744                                  * trim the existing memory map.
745                                  */
746                                 unsigned long long mem_size;
747  
748                                 mem_size = memparse(from+4, &from);
749                                 limit_regions(mem_size);
750                                 userdef=1;
751                         }
752                 }
753
754                 else if (!memcmp(from, "memmap=", 7)) {
755                         if (to != command_line)
756                                 to--;
757                         if (!memcmp(from+7, "exactmap", 8)) {
758 #ifdef CONFIG_CRASH_DUMP
759                                 /* If we are doing a crash dump, we
760                                  * still need to know the real mem
761                                  * size before original memory map is
762                                  * reset.
763                                  */
764                                 find_max_pfn();
765                                 saved_max_pfn = max_pfn;
766 #endif
767                                 from += 8+7;
768                                 e820.nr_map = 0;
769                                 userdef = 1;
770                         } else {
771                                 /* If the user specifies memory size, we
772                                  * limit the BIOS-provided memory map to
773                                  * that size. exactmap can be used to specify
774                                  * the exact map. mem=number can be used to
775                                  * trim the existing memory map.
776                                  */
777                                 unsigned long long start_at, mem_size;
778  
779                                 mem_size = memparse(from+7, &from);
780                                 if (*from == '@') {
781                                         start_at = memparse(from+1, &from);
782                                         add_memory_region(start_at, mem_size, E820_RAM);
783                                 } else if (*from == '#') {
784                                         start_at = memparse(from+1, &from);
785                                         add_memory_region(start_at, mem_size, E820_ACPI);
786                                 } else if (*from == '$') {
787                                         start_at = memparse(from+1, &from);
788                                         add_memory_region(start_at, mem_size, E820_RESERVED);
789                                 } else {
790                                         limit_regions(mem_size);
791                                         userdef=1;
792                                 }
793                         }
794                 }
795
796                 else if (!memcmp(from, "noexec=", 7))
797                         noexec_setup(from + 7);
798
799
800 #ifdef  CONFIG_X86_SMP
801                 /*
802                  * If the BIOS enumerates physical processors before logical,
803                  * maxcpus=N at enumeration-time can be used to disable HT.
804                  */
805                 else if (!memcmp(from, "maxcpus=", 8)) {
806                         extern unsigned int maxcpus;
807
808                         maxcpus = simple_strtoul(from + 8, NULL, 0);
809                 }
810 #endif
811
812 #ifdef CONFIG_ACPI
813                 /* "acpi=off" disables both ACPI table parsing and interpreter */
814                 else if (!memcmp(from, "acpi=off", 8)) {
815                         disable_acpi();
816                 }
817
818                 /* acpi=force to over-ride black-list */
819                 else if (!memcmp(from, "acpi=force", 10)) {
820                         acpi_force = 1;
821                         acpi_ht = 1;
822                         acpi_disabled = 0;
823                 }
824
825                 /* acpi=strict disables out-of-spec workarounds */
826                 else if (!memcmp(from, "acpi=strict", 11)) {
827                         acpi_strict = 1;
828                 }
829
830                 /* Limit ACPI just to boot-time to enable HT */
831                 else if (!memcmp(from, "acpi=ht", 7)) {
832                         if (!acpi_force)
833                                 disable_acpi();
834                         acpi_ht = 1;
835                 }
836                 
837                 /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */
838                 else if (!memcmp(from, "pci=noacpi", 10)) {
839                         acpi_disable_pci();
840                 }
841                 /* "acpi=noirq" disables ACPI interrupt routing */
842                 else if (!memcmp(from, "acpi=noirq", 10)) {
843                         acpi_noirq_set();
844                 }
845
846                 else if (!memcmp(from, "acpi_sci=edge", 13))
847                         acpi_sci_flags.trigger =  1;
848
849                 else if (!memcmp(from, "acpi_sci=level", 14))
850                         acpi_sci_flags.trigger = 3;
851
852                 else if (!memcmp(from, "acpi_sci=high", 13))
853                         acpi_sci_flags.polarity = 1;
854
855                 else if (!memcmp(from, "acpi_sci=low", 12))
856                         acpi_sci_flags.polarity = 3;
857
858 #ifdef CONFIG_X86_IO_APIC
859                 else if (!memcmp(from, "acpi_skip_timer_override", 24))
860                         acpi_skip_timer_override = 1;
861
862                 if (!memcmp(from, "disable_timer_pin_1", 19))
863                         disable_timer_pin_1 = 1;
864                 if (!memcmp(from, "enable_timer_pin_1", 18))
865                         disable_timer_pin_1 = -1;
866
867                 /* disable IO-APIC */
868                 else if (!memcmp(from, "noapic", 6))
869                         disable_ioapic_setup();
870 #endif /* CONFIG_X86_IO_APIC */
871 #endif /* CONFIG_ACPI */
872
873 #ifdef CONFIG_X86_LOCAL_APIC
874                 /* enable local APIC */
875                 else if (!memcmp(from, "lapic", 5))
876                         lapic_enable();
877
878                 /* disable local APIC */
879                 else if (!memcmp(from, "nolapic", 6))
880                         lapic_disable();
881 #endif /* CONFIG_X86_LOCAL_APIC */
882
883 #ifdef CONFIG_KEXEC
884                 /* crashkernel=size@addr specifies the location to reserve for
885                  * a crash kernel.  By reserving this memory we guarantee
886                  * that linux never set's it up as a DMA target.
887                  * Useful for holding code to do something appropriate
888                  * after a kernel panic.
889                  */
890                 else if (!memcmp(from, "crashkernel=", 12)) {
891                         unsigned long size, base;
892                         size = memparse(from+12, &from);
893                         if (*from == '@') {
894                                 base = memparse(from+1, &from);
895                                 /* FIXME: Do I want a sanity check
896                                  * to validate the memory range?
897                                  */
898                                 crashk_res.start = base;
899                                 crashk_res.end   = base + size - 1;
900                         }
901                 }
902 #endif
903 #ifdef CONFIG_PROC_VMCORE
904                 /* elfcorehdr= specifies the location of elf core header
905                  * stored by the crashed kernel.
906                  */
907                 else if (!memcmp(from, "elfcorehdr=", 11))
908                         elfcorehdr_addr = memparse(from+11, &from);
909 #endif
910
911                 /*
912                  * highmem=size forces highmem to be exactly 'size' bytes.
913                  * This works even on boxes that have no highmem otherwise.
914                  * This also works to reduce highmem size on bigger boxes.
915                  */
916                 else if (!memcmp(from, "highmem=", 8))
917                         highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT;
918         
919                 /*
920                  * vmalloc=size forces the vmalloc area to be exactly 'size'
921                  * bytes. This can be used to increase (or decrease) the
922                  * vmalloc area - the default is 128m.
923                  */
924                 else if (!memcmp(from, "vmalloc=", 8))
925                         __VMALLOC_RESERVE = memparse(from+8, &from);
926
927         next_char:
928                 c = *(from++);
929                 if (!c)
930                         break;
931                 if (COMMAND_LINE_SIZE <= ++len)
932                         break;
933                 *(to++) = c;
934         }
935         *to = '\0';
936         *cmdline_p = command_line;
937         if (userdef) {
938                 printk(KERN_INFO "user-defined physical RAM map:\n");
939                 print_memory_map("user");
940         }
941 }
942
943 /*
944  * Callback for efi_memory_walk.
945  */
946 static int __init
947 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
948 {
949         unsigned long *max_pfn = arg, pfn;
950
951         if (start < end) {
952                 pfn = PFN_UP(end -1);
953                 if (pfn > *max_pfn)
954                         *max_pfn = pfn;
955         }
956         return 0;
957 }
958
959 static int __init
960 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
961 {
962         memory_present(0, start, end);
963         return 0;
964 }
965
966  /*
967   * This function checks if the entire range <start,end> is mapped with type.
968   *
969   * Note: this function only works correct if the e820 table is sorted and
970   * not-overlapping, which is the case
971   */
972 int __init
973 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
974 {
975         u64 start = s;
976         u64 end = e;
977         int i;
978         for (i = 0; i < e820.nr_map; i++) {
979                 struct e820entry *ei = &e820.map[i];
980                 if (type && ei->type != type)
981                         continue;
982                 /* is the region (part) in overlap with the current region ?*/
983                 if (ei->addr >= end || ei->addr + ei->size <= start)
984                         continue;
985                 /* if the region is at the beginning of <start,end> we move
986                  * start to the end of the region since it's ok until there
987                  */
988                 if (ei->addr <= start)
989                         start = ei->addr + ei->size;
990                 /* if start is now at or beyond end, we're done, full
991                  * coverage */
992                 if (start >= end)
993                         return 1; /* we're done */
994         }
995         return 0;
996 }
997
998 /*
999  * Find the highest page frame number we have available
1000  */
1001 void __init find_max_pfn(void)
1002 {
1003         int i;
1004
1005         max_pfn = 0;
1006         if (efi_enabled) {
1007                 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
1008                 efi_memmap_walk(efi_memory_present_wrapper, NULL);
1009                 return;
1010         }
1011
1012         for (i = 0; i < e820.nr_map; i++) {
1013                 unsigned long start, end;
1014                 /* RAM? */
1015                 if (e820.map[i].type != E820_RAM)
1016                         continue;
1017                 start = PFN_UP(e820.map[i].addr);
1018                 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1019                 if (start >= end)
1020                         continue;
1021                 if (end > max_pfn)
1022                         max_pfn = end;
1023                 memory_present(0, start, end);
1024         }
1025 }
1026
1027 /*
1028  * Determine low and high memory ranges:
1029  */
1030 unsigned long __init find_max_low_pfn(void)
1031 {
1032         unsigned long max_low_pfn;
1033
1034         max_low_pfn = max_pfn;
1035         if (max_low_pfn > MAXMEM_PFN) {
1036                 if (highmem_pages == -1)
1037                         highmem_pages = max_pfn - MAXMEM_PFN;
1038                 if (highmem_pages + MAXMEM_PFN < max_pfn)
1039                         max_pfn = MAXMEM_PFN + highmem_pages;
1040                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
1041                         printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages));
1042                         highmem_pages = 0;
1043                 }
1044                 max_low_pfn = MAXMEM_PFN;
1045 #ifndef CONFIG_HIGHMEM
1046                 /* Maximum memory usable is what is directly addressable */
1047                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
1048                                         MAXMEM>>20);
1049                 if (max_pfn > MAX_NONPAE_PFN)
1050                         printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1051                 else
1052                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
1053                 max_pfn = MAXMEM_PFN;
1054 #else /* !CONFIG_HIGHMEM */
1055 #ifndef CONFIG_X86_PAE
1056                 if (max_pfn > MAX_NONPAE_PFN) {
1057                         max_pfn = MAX_NONPAE_PFN;
1058                         printk(KERN_WARNING "Warning only 4GB will be used.\n");
1059                         printk(KERN_WARNING "Use a PAE enabled kernel.\n");
1060                 }
1061 #endif /* !CONFIG_X86_PAE */
1062 #endif /* !CONFIG_HIGHMEM */
1063         } else {
1064                 if (highmem_pages == -1)
1065                         highmem_pages = 0;
1066 #ifdef CONFIG_HIGHMEM
1067                 if (highmem_pages >= max_pfn) {
1068                         printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
1069                         highmem_pages = 0;
1070                 }
1071                 if (highmem_pages) {
1072                         if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){
1073                                 printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages));
1074                                 highmem_pages = 0;
1075                         }
1076                         max_low_pfn -= highmem_pages;
1077                 }
1078 #else
1079                 if (highmem_pages)
1080                         printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
1081 #endif
1082         }
1083         return max_low_pfn;
1084 }
1085
1086 /*
1087  * Free all available memory for boot time allocation.  Used
1088  * as a callback function by efi_memory_walk()
1089  */
1090
1091 static int __init
1092 free_available_memory(unsigned long start, unsigned long end, void *arg)
1093 {
1094         /* check max_low_pfn */
1095         if (start >= (max_low_pfn << PAGE_SHIFT))
1096                 return 0;
1097         if (end >= (max_low_pfn << PAGE_SHIFT))
1098                 end = max_low_pfn << PAGE_SHIFT;
1099         if (start < end)
1100                 free_bootmem(start, end - start);
1101
1102         return 0;
1103 }
1104 /*
1105  * Register fully available low RAM pages with the bootmem allocator.
1106  */
1107 static void __init register_bootmem_low_pages(unsigned long max_low_pfn)
1108 {
1109         int i;
1110
1111         if (efi_enabled) {
1112                 efi_memmap_walk(free_available_memory, NULL);
1113                 return;
1114         }
1115         for (i = 0; i < e820.nr_map; i++) {
1116                 unsigned long curr_pfn, last_pfn, size;
1117                 /*
1118                  * Reserve usable low memory
1119                  */
1120                 if (e820.map[i].type != E820_RAM)
1121                         continue;
1122                 /*
1123                  * We are rounding up the start address of usable memory:
1124                  */
1125                 curr_pfn = PFN_UP(e820.map[i].addr);
1126                 if (curr_pfn >= max_low_pfn)
1127                         continue;
1128                 /*
1129                  * ... and at the end of the usable range downwards:
1130                  */
1131                 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
1132
1133                 if (last_pfn > max_low_pfn)
1134                         last_pfn = max_low_pfn;
1135
1136                 /*
1137                  * .. finally, did all the rounding and playing
1138                  * around just make the area go away?
1139                  */
1140                 if (last_pfn <= curr_pfn)
1141                         continue;
1142
1143                 size = last_pfn - curr_pfn;
1144                 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
1145         }
1146 }
1147
1148 /*
1149  * workaround for Dell systems that neglect to reserve EBDA
1150  */
1151 static void __init reserve_ebda_region(void)
1152 {
1153         unsigned int addr;
1154         addr = get_bios_ebda();
1155         if (addr)
1156                 reserve_bootmem(addr, PAGE_SIZE);       
1157 }
1158
1159 #ifndef CONFIG_NEED_MULTIPLE_NODES
1160 void __init setup_bootmem_allocator(void);
1161 static unsigned long __init setup_memory(void)
1162 {
1163         /*
1164          * partially used pages are not usable - thus
1165          * we are rounding upwards:
1166          */
1167         min_low_pfn = PFN_UP(init_pg_tables_end);
1168
1169         find_max_pfn();
1170
1171         max_low_pfn = find_max_low_pfn();
1172
1173 #ifdef CONFIG_HIGHMEM
1174         highstart_pfn = highend_pfn = max_pfn;
1175         if (max_pfn > max_low_pfn) {
1176                 highstart_pfn = max_low_pfn;
1177         }
1178         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
1179                 pages_to_mb(highend_pfn - highstart_pfn));
1180 #endif
1181         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
1182                         pages_to_mb(max_low_pfn));
1183
1184         setup_bootmem_allocator();
1185
1186         return max_low_pfn;
1187 }
1188
1189 void __init zone_sizes_init(void)
1190 {
1191         unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
1192         unsigned int max_dma, low;
1193
1194         max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
1195         low = max_low_pfn;
1196
1197         if (low < max_dma)
1198                 zones_size[ZONE_DMA] = low;
1199         else {
1200                 zones_size[ZONE_DMA] = max_dma;
1201                 zones_size[ZONE_NORMAL] = low - max_dma;
1202 #ifdef CONFIG_HIGHMEM
1203                 zones_size[ZONE_HIGHMEM] = highend_pfn - low;
1204 #endif
1205         }
1206         free_area_init(zones_size);
1207 }
1208 #else
1209 extern unsigned long __init setup_memory(void);
1210 extern void zone_sizes_init(void);
1211 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
1212
1213 void __init setup_bootmem_allocator(void)
1214 {
1215         unsigned long bootmap_size;
1216         /*
1217          * Initialize the boot-time allocator (with low memory only):
1218          */
1219         bootmap_size = init_bootmem(min_low_pfn, max_low_pfn);
1220
1221         register_bootmem_low_pages(max_low_pfn);
1222
1223         /*
1224          * Reserve the bootmem bitmap itself as well. We do this in two
1225          * steps (first step was init_bootmem()) because this catches
1226          * the (very unlikely) case of us accidentally initializing the
1227          * bootmem allocator with an invalid RAM area.
1228          */
1229         reserve_bootmem(__PHYSICAL_START, (PFN_PHYS(min_low_pfn) +
1230                          bootmap_size + PAGE_SIZE-1) - (__PHYSICAL_START));
1231
1232         /*
1233          * reserve physical page 0 - it's a special BIOS page on many boxes,
1234          * enabling clean reboots, SMP operation, laptop functions.
1235          */
1236         reserve_bootmem(0, PAGE_SIZE);
1237
1238         /* reserve EBDA region, it's a 4K region */
1239         reserve_ebda_region();
1240
1241     /* could be an AMD 768MPX chipset. Reserve a page  before VGA to prevent
1242        PCI prefetch into it (errata #56). Usually the page is reserved anyways,
1243        unless you have no PS/2 mouse plugged in. */
1244         if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
1245             boot_cpu_data.x86 == 6)
1246              reserve_bootmem(0xa0000 - 4096, 4096);
1247
1248 #ifdef CONFIG_SMP
1249         /*
1250          * But first pinch a few for the stack/trampoline stuff
1251          * FIXME: Don't need the extra page at 4K, but need to fix
1252          * trampoline before removing it. (see the GDT stuff)
1253          */
1254         reserve_bootmem(PAGE_SIZE, PAGE_SIZE);
1255 #endif
1256 #ifdef CONFIG_ACPI_SLEEP
1257         /*
1258          * Reserve low memory region for sleep support.
1259          */
1260         acpi_reserve_bootmem();
1261 #endif
1262 #ifdef CONFIG_X86_FIND_SMP_CONFIG
1263         /*
1264          * Find and reserve possible boot-time SMP configuration:
1265          */
1266         find_smp_config();
1267 #endif
1268
1269 #ifdef CONFIG_BLK_DEV_INITRD
1270         if (LOADER_TYPE && INITRD_START) {
1271                 if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) {
1272                         reserve_bootmem(INITRD_START, INITRD_SIZE);
1273                         initrd_start =
1274                                 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
1275                         initrd_end = initrd_start+INITRD_SIZE;
1276                 }
1277                 else {
1278                         printk(KERN_ERR "initrd extends beyond end of memory "
1279                             "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
1280                             INITRD_START + INITRD_SIZE,
1281                             max_low_pfn << PAGE_SHIFT);
1282                         initrd_start = 0;
1283                 }
1284         }
1285 #endif
1286 #ifdef CONFIG_KEXEC
1287         if (crashk_res.start != crashk_res.end)
1288                 reserve_bootmem(crashk_res.start,
1289                         crashk_res.end - crashk_res.start + 1);
1290 #endif
1291 }
1292
1293 /*
1294  * The node 0 pgdat is initialized before all of these because
1295  * it's needed for bootmem.  node>0 pgdats have their virtual
1296  * space allocated before the pagetables are in place to access
1297  * them, so they can't be cleared then.
1298  *
1299  * This should all compile down to nothing when NUMA is off.
1300  */
1301 void __init remapped_pgdat_init(void)
1302 {
1303         int nid;
1304
1305         for_each_online_node(nid) {
1306                 if (nid != 0)
1307                         memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
1308         }
1309 }
1310
1311 /*
1312  * Request address space for all standard RAM and ROM resources
1313  * and also for regions reported as reserved by the e820.
1314  */
1315 static void __init
1316 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
1317 {
1318         int i;
1319
1320         probe_roms();
1321         for (i = 0; i < e820.nr_map; i++) {
1322                 struct resource *res;
1323                 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
1324                         continue;
1325                 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
1326                 switch (e820.map[i].type) {
1327                 case E820_RAM:  res->name = "System RAM"; break;
1328                 case E820_ACPI: res->name = "ACPI Tables"; break;
1329                 case E820_NVS:  res->name = "ACPI Non-volatile Storage"; break;
1330                 default:        res->name = "reserved";
1331                 }
1332                 res->start = e820.map[i].addr;
1333                 res->end = res->start + e820.map[i].size - 1;
1334                 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1335                 request_resource(&iomem_resource, res);
1336                 if (e820.map[i].type == E820_RAM) {
1337                         /*
1338                          *  We don't know which RAM region contains kernel data,
1339                          *  so we try it repeatedly and let the resource manager
1340                          *  test it.
1341                          */
1342                         request_resource(res, code_resource);
1343                         request_resource(res, data_resource);
1344 #ifdef CONFIG_KEXEC
1345                         request_resource(res, &crashk_res);
1346 #endif
1347                 }
1348         }
1349 }
1350
1351 /*
1352  * Request address space for all standard resources
1353  *
1354  * This is called just before pcibios_init(), which is also a
1355  * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
1356  */
1357 static int __init request_standard_resources(void)
1358 {
1359         int i;
1360
1361         printk("Setting up standard PCI resources\n");
1362         if (efi_enabled)
1363                 efi_initialize_iomem_resources(&code_resource, &data_resource);
1364         else
1365                 legacy_init_iomem_resources(&code_resource, &data_resource);
1366
1367         /* EFI systems may still have VGA */
1368         request_resource(&iomem_resource, &video_ram_resource);
1369
1370         /* request I/O space for devices used on all i[345]86 PCs */
1371         for (i = 0; i < STANDARD_IO_RESOURCES; i++)
1372                 request_resource(&ioport_resource, &standard_io_resources[i]);
1373         return 0;
1374 }
1375
1376 subsys_initcall(request_standard_resources);
1377
1378 static void __init register_memory(void)
1379 {
1380         unsigned long gapstart, gapsize, round;
1381         unsigned long long last;
1382         int i;
1383
1384         /*
1385          * Search for the bigest gap in the low 32 bits of the e820
1386          * memory space.
1387          */
1388         last = 0x100000000ull;
1389         gapstart = 0x10000000;
1390         gapsize = 0x400000;
1391         i = e820.nr_map;
1392         while (--i >= 0) {
1393                 unsigned long long start = e820.map[i].addr;
1394                 unsigned long long end = start + e820.map[i].size;
1395
1396                 /*
1397                  * Since "last" is at most 4GB, we know we'll
1398                  * fit in 32 bits if this condition is true
1399                  */
1400                 if (last > end) {
1401                         unsigned long gap = last - end;
1402
1403                         if (gap > gapsize) {
1404                                 gapsize = gap;
1405                                 gapstart = end;
1406                         }
1407                 }
1408                 if (start < last)
1409                         last = start;
1410         }
1411
1412         /*
1413          * See how much we want to round up: start off with
1414          * rounding to the next 1MB area.
1415          */
1416         round = 0x100000;
1417         while ((gapsize >> 4) > round)
1418                 round += round;
1419         /* Fun with two's complement */
1420         pci_mem_start = (gapstart + round) & -round;
1421
1422         printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
1423                 pci_mem_start, gapstart, gapsize);
1424 }
1425
1426 static char * __init machine_specific_memory_setup(void);
1427
1428 #ifdef CONFIG_MCA
1429 static void set_mca_bus(int x)
1430 {
1431         MCA_bus = x;
1432 }
1433 #else
1434 static void set_mca_bus(int x) { }
1435 #endif
1436
1437 /*
1438  * Determine if we were loaded by an EFI loader.  If so, then we have also been
1439  * passed the efi memmap, systab, etc., so we should use these data structures
1440  * for initialization.  Note, the efi init code path is determined by the
1441  * global efi_enabled. This allows the same kernel image to be used on existing
1442  * systems (with a traditional BIOS) as well as on EFI systems.
1443  */
1444 void __init setup_arch(char **cmdline_p)
1445 {
1446         unsigned long max_low_pfn;
1447
1448         memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
1449         pre_setup_arch_hook();
1450         early_cpu_init();
1451
1452         /*
1453          * FIXME: This isn't an official loader_type right
1454          * now but does currently work with elilo.
1455          * If we were configured as an EFI kernel, check to make
1456          * sure that we were loaded correctly from elilo and that
1457          * the system table is valid.  If not, then initialize normally.
1458          */
1459 #ifdef CONFIG_EFI
1460         if ((LOADER_TYPE == 0x50) && EFI_SYSTAB)
1461                 efi_enabled = 1;
1462 #endif
1463
1464         ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
1465         drive_info = DRIVE_INFO;
1466         screen_info = SCREEN_INFO;
1467         edid_info = EDID_INFO;
1468         apm_info.bios = APM_BIOS_INFO;
1469         ist_info = IST_INFO;
1470         saved_videomode = VIDEO_MODE;
1471         if( SYS_DESC_TABLE.length != 0 ) {
1472                 set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2);
1473                 machine_id = SYS_DESC_TABLE.table[0];
1474                 machine_submodel_id = SYS_DESC_TABLE.table[1];
1475                 BIOS_revision = SYS_DESC_TABLE.table[2];
1476         }
1477         bootloader_type = LOADER_TYPE;
1478
1479 #ifdef CONFIG_BLK_DEV_RAM
1480         rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
1481         rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
1482         rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
1483 #endif
1484         ARCH_SETUP
1485         if (efi_enabled)
1486                 efi_init();
1487         else {
1488                 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1489                 print_memory_map(machine_specific_memory_setup());
1490         }
1491
1492         copy_edd();
1493
1494         if (!MOUNT_ROOT_RDONLY)
1495                 root_mountflags &= ~MS_RDONLY;
1496         init_mm.start_code = (unsigned long) _text;
1497         init_mm.end_code = (unsigned long) _etext;
1498         init_mm.end_data = (unsigned long) _edata;
1499         init_mm.brk = init_pg_tables_end + PAGE_OFFSET;
1500
1501         code_resource.start = virt_to_phys(_text);
1502         code_resource.end = virt_to_phys(_etext)-1;
1503         data_resource.start = virt_to_phys(_etext);
1504         data_resource.end = virt_to_phys(_edata)-1;
1505
1506         parse_cmdline_early(cmdline_p);
1507
1508 #ifdef CONFIG_EARLY_PRINTK
1509         {
1510                 char *s = strstr(*cmdline_p, "earlyprintk=");
1511                 if (s) {
1512                         setup_early_printk(strchr(s, '=') + 1);
1513                         printk("early console enabled\n");
1514                 }
1515         }
1516 #endif
1517
1518         max_low_pfn = setup_memory();
1519
1520         /*
1521          * NOTE: before this point _nobody_ is allowed to allocate
1522          * any memory using the bootmem allocator.  Although the
1523          * alloctor is now initialised only the first 8Mb of the kernel
1524          * virtual address space has been mapped.  All allocations before
1525          * paging_init() has completed must use the alloc_bootmem_low_pages()
1526          * variant (which allocates DMA'able memory) and care must be taken
1527          * not to exceed the 8Mb limit.
1528          */
1529
1530 #ifdef CONFIG_SMP
1531         smp_alloc_memory(); /* AP processor realmode stacks in low memory*/
1532 #endif
1533         paging_init();
1534         remapped_pgdat_init();
1535         sparse_init();
1536         zone_sizes_init();
1537
1538         /*
1539          * NOTE: at this point the bootmem allocator is fully available.
1540          */
1541
1542         dmi_scan_machine();
1543
1544 #ifdef CONFIG_X86_GENERICARCH
1545         generic_apic_probe(*cmdline_p);
1546 #endif  
1547         if (efi_enabled)
1548                 efi_map_memmap();
1549
1550 #ifdef CONFIG_ACPI
1551         /*
1552          * Parse the ACPI tables for possible boot-time SMP configuration.
1553          */
1554         acpi_boot_table_init();
1555 #endif
1556
1557 #ifdef CONFIG_X86_IO_APIC
1558         check_acpi_pci();       /* Checks more than just ACPI actually */
1559 #endif
1560
1561 #ifdef CONFIG_ACPI
1562         acpi_boot_init();
1563
1564 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PC)
1565         if (def_to_bigsmp)
1566                 printk(KERN_WARNING "More than 8 CPUs detected and "
1567                         "CONFIG_X86_PC cannot handle it.\nUse "
1568                         "CONFIG_X86_GENERICARCH or CONFIG_X86_BIGSMP.\n");
1569 #endif
1570 #endif
1571 #ifdef CONFIG_X86_LOCAL_APIC
1572         if (smp_found_config)
1573                 get_smp_config();
1574 #endif
1575
1576         register_memory();
1577
1578 #ifdef CONFIG_VT
1579 #if defined(CONFIG_VGA_CONSOLE)
1580         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1581                 conswitchp = &vga_con;
1582 #elif defined(CONFIG_DUMMY_CONSOLE)
1583         conswitchp = &dummy_con;
1584 #endif
1585 #endif
1586 }
1587
1588 static __init int add_pcspkr(void)
1589 {
1590         struct platform_device *pd;
1591         int ret;
1592
1593         pd = platform_device_alloc("pcspkr", -1);
1594         if (!pd)
1595                 return -ENOMEM;
1596
1597         ret = platform_device_add(pd);
1598         if (ret)
1599                 platform_device_put(pd);
1600
1601         return ret;
1602 }
1603 device_initcall(add_pcspkr);
1604
1605 #include "setup_arch_post.h"
1606 /*
1607  * Local Variables:
1608  * mode:c
1609  * c-file-style:"k&r"
1610  * c-basic-offset:8
1611  * End:
1612  */