]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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/sched.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/tty.h> | |
29 | #include <linux/ioport.h> | |
30 | #include <linux/acpi.h> | |
31 | #include <linux/apm_bios.h> | |
32 | #include <linux/initrd.h> | |
33 | #include <linux/bootmem.h> | |
34 | #include <linux/seq_file.h> | |
35 | #include <linux/console.h> | |
36 | #include <linux/mca.h> | |
37 | #include <linux/root_dev.h> | |
38 | #include <linux/highmem.h> | |
39 | #include <linux/module.h> | |
40 | #include <linux/efi.h> | |
41 | #include <linux/init.h> | |
42 | #include <linux/edd.h> | |
43 | #include <linux/nodemask.h> | |
44 | #include <video/edid.h> | |
45 | #include <asm/e820.h> | |
46 | #include <asm/mpspec.h> | |
47 | #include <asm/setup.h> | |
48 | #include <asm/arch_hooks.h> | |
49 | #include <asm/sections.h> | |
50 | #include <asm/io_apic.h> | |
51 | #include <asm/ist.h> | |
52 | #include <asm/io.h> | |
53 | #include "setup_arch_pre.h" | |
54 | #include <bios_ebda.h> | |
55 | ||
56 | /* This value is set up by the early boot code to point to the value | |
57 | immediately after the boot time page tables. It contains a *physical* | |
58 | address, and must not be in the .bss segment! */ | |
59 | unsigned long init_pg_tables_end __initdata = ~0UL; | |
60 | ||
61 | int disable_pse __initdata = 0; | |
62 | ||
63 | /* | |
64 | * Machine setup.. | |
65 | */ | |
66 | ||
67 | #ifdef CONFIG_EFI | |
68 | int efi_enabled = 0; | |
69 | EXPORT_SYMBOL(efi_enabled); | |
70 | #endif | |
71 | ||
72 | /* cpu data as detected by the assembly code in head.S */ | |
73 | struct cpuinfo_x86 new_cpu_data __initdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 }; | |
74 | /* common cpu data for all cpus */ | |
75 | struct cpuinfo_x86 boot_cpu_data = { 0, 0, 0, 0, -1, 1, 0, 0, -1 }; | |
76 | ||
77 | unsigned long mmu_cr4_features; | |
78 | ||
79 | #ifdef CONFIG_ACPI_INTERPRETER | |
80 | int acpi_disabled = 0; | |
81 | #else | |
82 | int acpi_disabled = 1; | |
83 | #endif | |
84 | EXPORT_SYMBOL(acpi_disabled); | |
85 | ||
86 | #ifdef CONFIG_ACPI_BOOT | |
87 | int __initdata acpi_force = 0; | |
88 | extern acpi_interrupt_flags acpi_sci_flags; | |
89 | #endif | |
90 | ||
91 | /* for MCA, but anyone else can use it if they want */ | |
92 | unsigned int machine_id; | |
93 | unsigned int machine_submodel_id; | |
94 | unsigned int BIOS_revision; | |
95 | unsigned int mca_pentium_flag; | |
96 | ||
97 | /* For PCI or other memory-mapped resources */ | |
98 | unsigned long pci_mem_start = 0x10000000; | |
99 | ||
100 | /* Boot loader ID as an integer, for the benefit of proc_dointvec */ | |
101 | int bootloader_type; | |
102 | ||
103 | /* user-defined highmem size */ | |
104 | static unsigned int highmem_pages = -1; | |
105 | ||
106 | /* | |
107 | * Setup options | |
108 | */ | |
109 | struct drive_info_struct { char dummy[32]; } drive_info; | |
110 | struct screen_info screen_info; | |
111 | struct apm_info apm_info; | |
112 | struct sys_desc_table_struct { | |
113 | unsigned short length; | |
114 | unsigned char table[0]; | |
115 | }; | |
116 | struct edid_info edid_info; | |
117 | struct ist_info ist_info; | |
118 | struct e820map e820; | |
119 | ||
120 | extern void early_cpu_init(void); | |
121 | extern void dmi_scan_machine(void); | |
122 | extern void generic_apic_probe(char *); | |
123 | extern int root_mountflags; | |
124 | ||
125 | unsigned long saved_videomode; | |
126 | ||
127 | #define RAMDISK_IMAGE_START_MASK 0x07FF | |
128 | #define RAMDISK_PROMPT_FLAG 0x8000 | |
129 | #define RAMDISK_LOAD_FLAG 0x4000 | |
130 | ||
131 | static char command_line[COMMAND_LINE_SIZE]; | |
132 | ||
133 | unsigned char __initdata boot_params[PARAM_SIZE]; | |
134 | ||
135 | static struct resource data_resource = { | |
136 | .name = "Kernel data", | |
137 | .start = 0, | |
138 | .end = 0, | |
139 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM | |
140 | }; | |
141 | ||
142 | static struct resource code_resource = { | |
143 | .name = "Kernel code", | |
144 | .start = 0, | |
145 | .end = 0, | |
146 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM | |
147 | }; | |
148 | ||
149 | static struct resource system_rom_resource = { | |
150 | .name = "System ROM", | |
151 | .start = 0xf0000, | |
152 | .end = 0xfffff, | |
153 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
154 | }; | |
155 | ||
156 | static struct resource extension_rom_resource = { | |
157 | .name = "Extension ROM", | |
158 | .start = 0xe0000, | |
159 | .end = 0xeffff, | |
160 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
161 | }; | |
162 | ||
163 | static struct resource adapter_rom_resources[] = { { | |
164 | .name = "Adapter ROM", | |
165 | .start = 0xc8000, | |
166 | .end = 0, | |
167 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
168 | }, { | |
169 | .name = "Adapter ROM", | |
170 | .start = 0, | |
171 | .end = 0, | |
172 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
173 | }, { | |
174 | .name = "Adapter ROM", | |
175 | .start = 0, | |
176 | .end = 0, | |
177 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
178 | }, { | |
179 | .name = "Adapter ROM", | |
180 | .start = 0, | |
181 | .end = 0, | |
182 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
183 | }, { | |
184 | .name = "Adapter ROM", | |
185 | .start = 0, | |
186 | .end = 0, | |
187 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
188 | }, { | |
189 | .name = "Adapter ROM", | |
190 | .start = 0, | |
191 | .end = 0, | |
192 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
193 | } }; | |
194 | ||
195 | #define ADAPTER_ROM_RESOURCES \ | |
196 | (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0]) | |
197 | ||
198 | static struct resource video_rom_resource = { | |
199 | .name = "Video ROM", | |
200 | .start = 0xc0000, | |
201 | .end = 0xc7fff, | |
202 | .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM | |
203 | }; | |
204 | ||
205 | static struct resource video_ram_resource = { | |
206 | .name = "Video RAM area", | |
207 | .start = 0xa0000, | |
208 | .end = 0xbffff, | |
209 | .flags = IORESOURCE_BUSY | IORESOURCE_MEM | |
210 | }; | |
211 | ||
212 | static struct resource standard_io_resources[] = { { | |
213 | .name = "dma1", | |
214 | .start = 0x0000, | |
215 | .end = 0x001f, | |
216 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
217 | }, { | |
218 | .name = "pic1", | |
219 | .start = 0x0020, | |
220 | .end = 0x0021, | |
221 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
222 | }, { | |
223 | .name = "timer0", | |
224 | .start = 0x0040, | |
225 | .end = 0x0043, | |
226 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
227 | }, { | |
228 | .name = "timer1", | |
229 | .start = 0x0050, | |
230 | .end = 0x0053, | |
231 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
232 | }, { | |
233 | .name = "keyboard", | |
234 | .start = 0x0060, | |
235 | .end = 0x006f, | |
236 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
237 | }, { | |
238 | .name = "dma page reg", | |
239 | .start = 0x0080, | |
240 | .end = 0x008f, | |
241 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
242 | }, { | |
243 | .name = "pic2", | |
244 | .start = 0x00a0, | |
245 | .end = 0x00a1, | |
246 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
247 | }, { | |
248 | .name = "dma2", | |
249 | .start = 0x00c0, | |
250 | .end = 0x00df, | |
251 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
252 | }, { | |
253 | .name = "fpu", | |
254 | .start = 0x00f0, | |
255 | .end = 0x00ff, | |
256 | .flags = IORESOURCE_BUSY | IORESOURCE_IO | |
257 | } }; | |
258 | ||
259 | #define STANDARD_IO_RESOURCES \ | |
260 | (sizeof standard_io_resources / sizeof standard_io_resources[0]) | |
261 | ||
262 | #define romsignature(x) (*(unsigned short *)(x) == 0xaa55) | |
263 | ||
264 | static int __init romchecksum(unsigned char *rom, unsigned long length) | |
265 | { | |
266 | unsigned char *p, sum = 0; | |
267 | ||
268 | for (p = rom; p < rom + length; p++) | |
269 | sum += *p; | |
270 | return sum == 0; | |
271 | } | |
272 | ||
273 | static void __init probe_roms(void) | |
274 | { | |
275 | unsigned long start, length, upper; | |
276 | unsigned char *rom; | |
277 | int i; | |
278 | ||
279 | /* video rom */ | |
280 | upper = adapter_rom_resources[0].start; | |
281 | for (start = video_rom_resource.start; start < upper; start += 2048) { | |
282 | rom = isa_bus_to_virt(start); | |
283 | if (!romsignature(rom)) | |
284 | continue; | |
285 | ||
286 | video_rom_resource.start = start; | |
287 | ||
288 | /* 0 < length <= 0x7f * 512, historically */ | |
289 | length = rom[2] * 512; | |
290 | ||
291 | /* if checksum okay, trust length byte */ | |
292 | if (length && romchecksum(rom, length)) | |
293 | video_rom_resource.end = start + length - 1; | |
294 | ||
295 | request_resource(&iomem_resource, &video_rom_resource); | |
296 | break; | |
297 | } | |
298 | ||
299 | start = (video_rom_resource.end + 1 + 2047) & ~2047UL; | |
300 | if (start < upper) | |
301 | start = upper; | |
302 | ||
303 | /* system rom */ | |
304 | request_resource(&iomem_resource, &system_rom_resource); | |
305 | upper = system_rom_resource.start; | |
306 | ||
307 | /* check for extension rom (ignore length byte!) */ | |
308 | rom = isa_bus_to_virt(extension_rom_resource.start); | |
309 | if (romsignature(rom)) { | |
310 | length = extension_rom_resource.end - extension_rom_resource.start + 1; | |
311 | if (romchecksum(rom, length)) { | |
312 | request_resource(&iomem_resource, &extension_rom_resource); | |
313 | upper = extension_rom_resource.start; | |
314 | } | |
315 | } | |
316 | ||
317 | /* check for adapter roms on 2k boundaries */ | |
318 | for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) { | |
319 | rom = isa_bus_to_virt(start); | |
320 | if (!romsignature(rom)) | |
321 | continue; | |
322 | ||
323 | /* 0 < length <= 0x7f * 512, historically */ | |
324 | length = rom[2] * 512; | |
325 | ||
326 | /* but accept any length that fits if checksum okay */ | |
327 | if (!length || start + length > upper || !romchecksum(rom, length)) | |
328 | continue; | |
329 | ||
330 | adapter_rom_resources[i].start = start; | |
331 | adapter_rom_resources[i].end = start + length - 1; | |
332 | request_resource(&iomem_resource, &adapter_rom_resources[i]); | |
333 | ||
334 | start = adapter_rom_resources[i++].end & ~2047UL; | |
335 | } | |
336 | } | |
337 | ||
338 | static void __init limit_regions(unsigned long long size) | |
339 | { | |
340 | unsigned long long current_addr = 0; | |
341 | int i; | |
342 | ||
343 | if (efi_enabled) { | |
344 | for (i = 0; i < memmap.nr_map; i++) { | |
345 | current_addr = memmap.map[i].phys_addr + | |
346 | (memmap.map[i].num_pages << 12); | |
347 | if (memmap.map[i].type == EFI_CONVENTIONAL_MEMORY) { | |
348 | if (current_addr >= size) { | |
349 | memmap.map[i].num_pages -= | |
350 | (((current_addr-size) + PAGE_SIZE-1) >> PAGE_SHIFT); | |
351 | memmap.nr_map = i + 1; | |
352 | return; | |
353 | } | |
354 | } | |
355 | } | |
356 | } | |
357 | for (i = 0; i < e820.nr_map; i++) { | |
358 | if (e820.map[i].type == E820_RAM) { | |
359 | current_addr = e820.map[i].addr + e820.map[i].size; | |
360 | if (current_addr >= size) { | |
361 | e820.map[i].size -= current_addr-size; | |
362 | e820.nr_map = i + 1; | |
363 | return; | |
364 | } | |
365 | } | |
366 | } | |
367 | } | |
368 | ||
369 | static void __init add_memory_region(unsigned long long start, | |
370 | unsigned long long size, int type) | |
371 | { | |
372 | int x; | |
373 | ||
374 | if (!efi_enabled) { | |
375 | x = e820.nr_map; | |
376 | ||
377 | if (x == E820MAX) { | |
378 | printk(KERN_ERR "Ooops! Too many entries in the memory map!\n"); | |
379 | return; | |
380 | } | |
381 | ||
382 | e820.map[x].addr = start; | |
383 | e820.map[x].size = size; | |
384 | e820.map[x].type = type; | |
385 | e820.nr_map++; | |
386 | } | |
387 | } /* add_memory_region */ | |
388 | ||
389 | #define E820_DEBUG 1 | |
390 | ||
391 | static void __init print_memory_map(char *who) | |
392 | { | |
393 | int i; | |
394 | ||
395 | for (i = 0; i < e820.nr_map; i++) { | |
396 | printk(" %s: %016Lx - %016Lx ", who, | |
397 | e820.map[i].addr, | |
398 | e820.map[i].addr + e820.map[i].size); | |
399 | switch (e820.map[i].type) { | |
400 | case E820_RAM: printk("(usable)\n"); | |
401 | break; | |
402 | case E820_RESERVED: | |
403 | printk("(reserved)\n"); | |
404 | break; | |
405 | case E820_ACPI: | |
406 | printk("(ACPI data)\n"); | |
407 | break; | |
408 | case E820_NVS: | |
409 | printk("(ACPI NVS)\n"); | |
410 | break; | |
411 | default: printk("type %lu\n", e820.map[i].type); | |
412 | break; | |
413 | } | |
414 | } | |
415 | } | |
416 | ||
417 | /* | |
418 | * Sanitize the BIOS e820 map. | |
419 | * | |
420 | * Some e820 responses include overlapping entries. The following | |
421 | * replaces the original e820 map with a new one, removing overlaps. | |
422 | * | |
423 | */ | |
424 | struct change_member { | |
425 | struct e820entry *pbios; /* pointer to original bios entry */ | |
426 | unsigned long long addr; /* address for this change point */ | |
427 | }; | |
428 | static struct change_member change_point_list[2*E820MAX] __initdata; | |
429 | static struct change_member *change_point[2*E820MAX] __initdata; | |
430 | static struct e820entry *overlap_list[E820MAX] __initdata; | |
431 | static struct e820entry new_bios[E820MAX] __initdata; | |
432 | ||
433 | static int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map) | |
434 | { | |
435 | struct change_member *change_tmp; | |
436 | unsigned long current_type, last_type; | |
437 | unsigned long long last_addr; | |
438 | int chgidx, still_changing; | |
439 | int overlap_entries; | |
440 | int new_bios_entry; | |
441 | int old_nr, new_nr, chg_nr; | |
442 | int i; | |
443 | ||
444 | /* | |
445 | Visually we're performing the following (1,2,3,4 = memory types)... | |
446 | ||
447 | Sample memory map (w/overlaps): | |
448 | ____22__________________ | |
449 | ______________________4_ | |
450 | ____1111________________ | |
451 | _44_____________________ | |
452 | 11111111________________ | |
453 | ____________________33__ | |
454 | ___________44___________ | |
455 | __________33333_________ | |
456 | ______________22________ | |
457 | ___________________2222_ | |
458 | _________111111111______ | |
459 | _____________________11_ | |
460 | _________________4______ | |
461 | ||
462 | Sanitized equivalent (no overlap): | |
463 | 1_______________________ | |
464 | _44_____________________ | |
465 | ___1____________________ | |
466 | ____22__________________ | |
467 | ______11________________ | |
468 | _________1______________ | |
469 | __________3_____________ | |
470 | ___________44___________ | |
471 | _____________33_________ | |
472 | _______________2________ | |
473 | ________________1_______ | |
474 | _________________4______ | |
475 | ___________________2____ | |
476 | ____________________33__ | |
477 | ______________________4_ | |
478 | */ | |
479 | ||
480 | /* if there's only one memory region, don't bother */ | |
481 | if (*pnr_map < 2) | |
482 | return -1; | |
483 | ||
484 | old_nr = *pnr_map; | |
485 | ||
486 | /* bail out if we find any unreasonable addresses in bios map */ | |
487 | for (i=0; i<old_nr; i++) | |
488 | if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) | |
489 | return -1; | |
490 | ||
491 | /* create pointers for initial change-point information (for sorting) */ | |
492 | for (i=0; i < 2*old_nr; i++) | |
493 | change_point[i] = &change_point_list[i]; | |
494 | ||
495 | /* record all known change-points (starting and ending addresses), | |
496 | omitting those that are for empty memory regions */ | |
497 | chgidx = 0; | |
498 | for (i=0; i < old_nr; i++) { | |
499 | if (biosmap[i].size != 0) { | |
500 | change_point[chgidx]->addr = biosmap[i].addr; | |
501 | change_point[chgidx++]->pbios = &biosmap[i]; | |
502 | change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size; | |
503 | change_point[chgidx++]->pbios = &biosmap[i]; | |
504 | } | |
505 | } | |
506 | chg_nr = chgidx; /* true number of change-points */ | |
507 | ||
508 | /* sort change-point list by memory addresses (low -> high) */ | |
509 | still_changing = 1; | |
510 | while (still_changing) { | |
511 | still_changing = 0; | |
512 | for (i=1; i < chg_nr; i++) { | |
513 | /* if <current_addr> > <last_addr>, swap */ | |
514 | /* or, if current=<start_addr> & last=<end_addr>, swap */ | |
515 | if ((change_point[i]->addr < change_point[i-1]->addr) || | |
516 | ((change_point[i]->addr == change_point[i-1]->addr) && | |
517 | (change_point[i]->addr == change_point[i]->pbios->addr) && | |
518 | (change_point[i-1]->addr != change_point[i-1]->pbios->addr)) | |
519 | ) | |
520 | { | |
521 | change_tmp = change_point[i]; | |
522 | change_point[i] = change_point[i-1]; | |
523 | change_point[i-1] = change_tmp; | |
524 | still_changing=1; | |
525 | } | |
526 | } | |
527 | } | |
528 | ||
529 | /* create a new bios memory map, removing overlaps */ | |
530 | overlap_entries=0; /* number of entries in the overlap table */ | |
531 | new_bios_entry=0; /* index for creating new bios map entries */ | |
532 | last_type = 0; /* start with undefined memory type */ | |
533 | last_addr = 0; /* start with 0 as last starting address */ | |
534 | /* loop through change-points, determining affect on the new bios map */ | |
535 | for (chgidx=0; chgidx < chg_nr; chgidx++) | |
536 | { | |
537 | /* keep track of all overlapping bios entries */ | |
538 | if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr) | |
539 | { | |
540 | /* add map entry to overlap list (> 1 entry implies an overlap) */ | |
541 | overlap_list[overlap_entries++]=change_point[chgidx]->pbios; | |
542 | } | |
543 | else | |
544 | { | |
545 | /* remove entry from list (order independent, so swap with last) */ | |
546 | for (i=0; i<overlap_entries; i++) | |
547 | { | |
548 | if (overlap_list[i] == change_point[chgidx]->pbios) | |
549 | overlap_list[i] = overlap_list[overlap_entries-1]; | |
550 | } | |
551 | overlap_entries--; | |
552 | } | |
553 | /* if there are overlapping entries, decide which "type" to use */ | |
554 | /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */ | |
555 | current_type = 0; | |
556 | for (i=0; i<overlap_entries; i++) | |
557 | if (overlap_list[i]->type > current_type) | |
558 | current_type = overlap_list[i]->type; | |
559 | /* continue building up new bios map based on this information */ | |
560 | if (current_type != last_type) { | |
561 | if (last_type != 0) { | |
562 | new_bios[new_bios_entry].size = | |
563 | change_point[chgidx]->addr - last_addr; | |
564 | /* move forward only if the new size was non-zero */ | |
565 | if (new_bios[new_bios_entry].size != 0) | |
566 | if (++new_bios_entry >= E820MAX) | |
567 | break; /* no more space left for new bios entries */ | |
568 | } | |
569 | if (current_type != 0) { | |
570 | new_bios[new_bios_entry].addr = change_point[chgidx]->addr; | |
571 | new_bios[new_bios_entry].type = current_type; | |
572 | last_addr=change_point[chgidx]->addr; | |
573 | } | |
574 | last_type = current_type; | |
575 | } | |
576 | } | |
577 | new_nr = new_bios_entry; /* retain count for new bios entries */ | |
578 | ||
579 | /* copy new bios mapping into original location */ | |
580 | memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry)); | |
581 | *pnr_map = new_nr; | |
582 | ||
583 | return 0; | |
584 | } | |
585 | ||
586 | /* | |
587 | * Copy the BIOS e820 map into a safe place. | |
588 | * | |
589 | * Sanity-check it while we're at it.. | |
590 | * | |
591 | * If we're lucky and live on a modern system, the setup code | |
592 | * will have given us a memory map that we can use to properly | |
593 | * set up memory. If we aren't, we'll fake a memory map. | |
594 | * | |
595 | * We check to see that the memory map contains at least 2 elements | |
596 | * before we'll use it, because the detection code in setup.S may | |
597 | * not be perfect and most every PC known to man has two memory | |
598 | * regions: one from 0 to 640k, and one from 1mb up. (The IBM | |
599 | * thinkpad 560x, for example, does not cooperate with the memory | |
600 | * detection code.) | |
601 | */ | |
602 | static int __init copy_e820_map(struct e820entry * biosmap, int nr_map) | |
603 | { | |
604 | /* Only one memory region (or negative)? Ignore it */ | |
605 | if (nr_map < 2) | |
606 | return -1; | |
607 | ||
608 | do { | |
609 | unsigned long long start = biosmap->addr; | |
610 | unsigned long long size = biosmap->size; | |
611 | unsigned long long end = start + size; | |
612 | unsigned long type = biosmap->type; | |
613 | ||
614 | /* Overflow in 64 bits? Ignore the memory map. */ | |
615 | if (start > end) | |
616 | return -1; | |
617 | ||
618 | /* | |
619 | * Some BIOSes claim RAM in the 640k - 1M region. | |
620 | * Not right. Fix it up. | |
621 | */ | |
622 | if (type == E820_RAM) { | |
623 | if (start < 0x100000ULL && end > 0xA0000ULL) { | |
624 | if (start < 0xA0000ULL) | |
625 | add_memory_region(start, 0xA0000ULL-start, type); | |
626 | if (end <= 0x100000ULL) | |
627 | continue; | |
628 | start = 0x100000ULL; | |
629 | size = end - start; | |
630 | } | |
631 | } | |
632 | add_memory_region(start, size, type); | |
633 | } while (biosmap++,--nr_map); | |
634 | return 0; | |
635 | } | |
636 | ||
637 | #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE) | |
638 | struct edd edd; | |
639 | #ifdef CONFIG_EDD_MODULE | |
640 | EXPORT_SYMBOL(edd); | |
641 | #endif | |
642 | /** | |
643 | * copy_edd() - Copy the BIOS EDD information | |
644 | * from boot_params into a safe place. | |
645 | * | |
646 | */ | |
647 | static inline void copy_edd(void) | |
648 | { | |
649 | memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature)); | |
650 | memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info)); | |
651 | edd.mbr_signature_nr = EDD_MBR_SIG_NR; | |
652 | edd.edd_info_nr = EDD_NR; | |
653 | } | |
654 | #else | |
655 | static inline void copy_edd(void) | |
656 | { | |
657 | } | |
658 | #endif | |
659 | ||
660 | /* | |
661 | * Do NOT EVER look at the BIOS memory size location. | |
662 | * It does not work on many machines. | |
663 | */ | |
664 | #define LOWMEMSIZE() (0x9f000) | |
665 | ||
666 | static void __init parse_cmdline_early (char ** cmdline_p) | |
667 | { | |
668 | char c = ' ', *to = command_line, *from = saved_command_line; | |
669 | int len = 0; | |
670 | int userdef = 0; | |
671 | ||
672 | /* Save unparsed command line copy for /proc/cmdline */ | |
673 | saved_command_line[COMMAND_LINE_SIZE-1] = '\0'; | |
674 | ||
675 | for (;;) { | |
676 | if (c != ' ') | |
677 | goto next_char; | |
678 | /* | |
679 | * "mem=nopentium" disables the 4MB page tables. | |
680 | * "mem=XXX[kKmM]" defines a memory region from HIGH_MEM | |
681 | * to <mem>, overriding the bios size. | |
682 | * "memmap=XXX[KkmM]@XXX[KkmM]" defines a memory region from | |
683 | * <start> to <start>+<mem>, overriding the bios size. | |
684 | * | |
685 | * HPA tells me bootloaders need to parse mem=, so no new | |
686 | * option should be mem= [also see Documentation/i386/boot.txt] | |
687 | */ | |
688 | if (!memcmp(from, "mem=", 4)) { | |
689 | if (to != command_line) | |
690 | to--; | |
691 | if (!memcmp(from+4, "nopentium", 9)) { | |
692 | from += 9+4; | |
693 | clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability); | |
694 | disable_pse = 1; | |
695 | } else { | |
696 | /* If the user specifies memory size, we | |
697 | * limit the BIOS-provided memory map to | |
698 | * that size. exactmap can be used to specify | |
699 | * the exact map. mem=number can be used to | |
700 | * trim the existing memory map. | |
701 | */ | |
702 | unsigned long long mem_size; | |
703 | ||
704 | mem_size = memparse(from+4, &from); | |
705 | limit_regions(mem_size); | |
706 | userdef=1; | |
707 | } | |
708 | } | |
709 | ||
710 | else if (!memcmp(from, "memmap=", 7)) { | |
711 | if (to != command_line) | |
712 | to--; | |
713 | if (!memcmp(from+7, "exactmap", 8)) { | |
714 | from += 8+7; | |
715 | e820.nr_map = 0; | |
716 | userdef = 1; | |
717 | } else { | |
718 | /* If the user specifies memory size, we | |
719 | * limit the BIOS-provided memory map to | |
720 | * that size. exactmap can be used to specify | |
721 | * the exact map. mem=number can be used to | |
722 | * trim the existing memory map. | |
723 | */ | |
724 | unsigned long long start_at, mem_size; | |
725 | ||
726 | mem_size = memparse(from+7, &from); | |
727 | if (*from == '@') { | |
728 | start_at = memparse(from+1, &from); | |
729 | add_memory_region(start_at, mem_size, E820_RAM); | |
730 | } else if (*from == '#') { | |
731 | start_at = memparse(from+1, &from); | |
732 | add_memory_region(start_at, mem_size, E820_ACPI); | |
733 | } else if (*from == '$') { | |
734 | start_at = memparse(from+1, &from); | |
735 | add_memory_region(start_at, mem_size, E820_RESERVED); | |
736 | } else { | |
737 | limit_regions(mem_size); | |
738 | userdef=1; | |
739 | } | |
740 | } | |
741 | } | |
742 | ||
743 | else if (!memcmp(from, "noexec=", 7)) | |
744 | noexec_setup(from + 7); | |
745 | ||
746 | ||
747 | #ifdef CONFIG_X86_SMP | |
748 | /* | |
749 | * If the BIOS enumerates physical processors before logical, | |
750 | * maxcpus=N at enumeration-time can be used to disable HT. | |
751 | */ | |
752 | else if (!memcmp(from, "maxcpus=", 8)) { | |
753 | extern unsigned int maxcpus; | |
754 | ||
755 | maxcpus = simple_strtoul(from + 8, NULL, 0); | |
756 | } | |
757 | #endif | |
758 | ||
759 | #ifdef CONFIG_ACPI_BOOT | |
760 | /* "acpi=off" disables both ACPI table parsing and interpreter */ | |
761 | else if (!memcmp(from, "acpi=off", 8)) { | |
762 | disable_acpi(); | |
763 | } | |
764 | ||
765 | /* acpi=force to over-ride black-list */ | |
766 | else if (!memcmp(from, "acpi=force", 10)) { | |
767 | acpi_force = 1; | |
768 | acpi_ht = 1; | |
769 | acpi_disabled = 0; | |
770 | } | |
771 | ||
772 | /* acpi=strict disables out-of-spec workarounds */ | |
773 | else if (!memcmp(from, "acpi=strict", 11)) { | |
774 | acpi_strict = 1; | |
775 | } | |
776 | ||
777 | /* Limit ACPI just to boot-time to enable HT */ | |
778 | else if (!memcmp(from, "acpi=ht", 7)) { | |
779 | if (!acpi_force) | |
780 | disable_acpi(); | |
781 | acpi_ht = 1; | |
782 | } | |
783 | ||
784 | /* "pci=noacpi" disable ACPI IRQ routing and PCI scan */ | |
785 | else if (!memcmp(from, "pci=noacpi", 10)) { | |
786 | acpi_disable_pci(); | |
787 | } | |
788 | /* "acpi=noirq" disables ACPI interrupt routing */ | |
789 | else if (!memcmp(from, "acpi=noirq", 10)) { | |
790 | acpi_noirq_set(); | |
791 | } | |
792 | ||
793 | else if (!memcmp(from, "acpi_sci=edge", 13)) | |
794 | acpi_sci_flags.trigger = 1; | |
795 | ||
796 | else if (!memcmp(from, "acpi_sci=level", 14)) | |
797 | acpi_sci_flags.trigger = 3; | |
798 | ||
799 | else if (!memcmp(from, "acpi_sci=high", 13)) | |
800 | acpi_sci_flags.polarity = 1; | |
801 | ||
802 | else if (!memcmp(from, "acpi_sci=low", 12)) | |
803 | acpi_sci_flags.polarity = 3; | |
804 | ||
805 | #ifdef CONFIG_X86_IO_APIC | |
806 | else if (!memcmp(from, "acpi_skip_timer_override", 24)) | |
807 | acpi_skip_timer_override = 1; | |
808 | #endif | |
809 | ||
810 | #ifdef CONFIG_X86_LOCAL_APIC | |
811 | /* disable IO-APIC */ | |
812 | else if (!memcmp(from, "noapic", 6)) | |
813 | disable_ioapic_setup(); | |
814 | #endif /* CONFIG_X86_LOCAL_APIC */ | |
815 | #endif /* CONFIG_ACPI_BOOT */ | |
816 | ||
817 | /* | |
818 | * highmem=size forces highmem to be exactly 'size' bytes. | |
819 | * This works even on boxes that have no highmem otherwise. | |
820 | * This also works to reduce highmem size on bigger boxes. | |
821 | */ | |
822 | else if (!memcmp(from, "highmem=", 8)) | |
823 | highmem_pages = memparse(from+8, &from) >> PAGE_SHIFT; | |
824 | ||
825 | /* | |
826 | * vmalloc=size forces the vmalloc area to be exactly 'size' | |
827 | * bytes. This can be used to increase (or decrease) the | |
828 | * vmalloc area - the default is 128m. | |
829 | */ | |
830 | else if (!memcmp(from, "vmalloc=", 8)) | |
831 | __VMALLOC_RESERVE = memparse(from+8, &from); | |
832 | ||
833 | next_char: | |
834 | c = *(from++); | |
835 | if (!c) | |
836 | break; | |
837 | if (COMMAND_LINE_SIZE <= ++len) | |
838 | break; | |
839 | *(to++) = c; | |
840 | } | |
841 | *to = '\0'; | |
842 | *cmdline_p = command_line; | |
843 | if (userdef) { | |
844 | printk(KERN_INFO "user-defined physical RAM map:\n"); | |
845 | print_memory_map("user"); | |
846 | } | |
847 | } | |
848 | ||
849 | /* | |
850 | * Callback for efi_memory_walk. | |
851 | */ | |
852 | static int __init | |
853 | efi_find_max_pfn(unsigned long start, unsigned long end, void *arg) | |
854 | { | |
855 | unsigned long *max_pfn = arg, pfn; | |
856 | ||
857 | if (start < end) { | |
858 | pfn = PFN_UP(end -1); | |
859 | if (pfn > *max_pfn) | |
860 | *max_pfn = pfn; | |
861 | } | |
862 | return 0; | |
863 | } | |
864 | ||
865 | ||
866 | /* | |
867 | * Find the highest page frame number we have available | |
868 | */ | |
869 | void __init find_max_pfn(void) | |
870 | { | |
871 | int i; | |
872 | ||
873 | max_pfn = 0; | |
874 | if (efi_enabled) { | |
875 | efi_memmap_walk(efi_find_max_pfn, &max_pfn); | |
876 | return; | |
877 | } | |
878 | ||
879 | for (i = 0; i < e820.nr_map; i++) { | |
880 | unsigned long start, end; | |
881 | /* RAM? */ | |
882 | if (e820.map[i].type != E820_RAM) | |
883 | continue; | |
884 | start = PFN_UP(e820.map[i].addr); | |
885 | end = PFN_DOWN(e820.map[i].addr + e820.map[i].size); | |
886 | if (start >= end) | |
887 | continue; | |
888 | if (end > max_pfn) | |
889 | max_pfn = end; | |
890 | } | |
891 | } | |
892 | ||
893 | /* | |
894 | * Determine low and high memory ranges: | |
895 | */ | |
896 | unsigned long __init find_max_low_pfn(void) | |
897 | { | |
898 | unsigned long max_low_pfn; | |
899 | ||
900 | max_low_pfn = max_pfn; | |
901 | if (max_low_pfn > MAXMEM_PFN) { | |
902 | if (highmem_pages == -1) | |
903 | highmem_pages = max_pfn - MAXMEM_PFN; | |
904 | if (highmem_pages + MAXMEM_PFN < max_pfn) | |
905 | max_pfn = MAXMEM_PFN + highmem_pages; | |
906 | if (highmem_pages + MAXMEM_PFN > max_pfn) { | |
907 | printk("only %luMB highmem pages available, ignoring highmem size of %uMB.\n", pages_to_mb(max_pfn - MAXMEM_PFN), pages_to_mb(highmem_pages)); | |
908 | highmem_pages = 0; | |
909 | } | |
910 | max_low_pfn = MAXMEM_PFN; | |
911 | #ifndef CONFIG_HIGHMEM | |
912 | /* Maximum memory usable is what is directly addressable */ | |
913 | printk(KERN_WARNING "Warning only %ldMB will be used.\n", | |
914 | MAXMEM>>20); | |
915 | if (max_pfn > MAX_NONPAE_PFN) | |
916 | printk(KERN_WARNING "Use a PAE enabled kernel.\n"); | |
917 | else | |
918 | printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n"); | |
919 | max_pfn = MAXMEM_PFN; | |
920 | #else /* !CONFIG_HIGHMEM */ | |
921 | #ifndef CONFIG_X86_PAE | |
922 | if (max_pfn > MAX_NONPAE_PFN) { | |
923 | max_pfn = MAX_NONPAE_PFN; | |
924 | printk(KERN_WARNING "Warning only 4GB will be used.\n"); | |
925 | printk(KERN_WARNING "Use a PAE enabled kernel.\n"); | |
926 | } | |
927 | #endif /* !CONFIG_X86_PAE */ | |
928 | #endif /* !CONFIG_HIGHMEM */ | |
929 | } else { | |
930 | if (highmem_pages == -1) | |
931 | highmem_pages = 0; | |
932 | #ifdef CONFIG_HIGHMEM | |
933 | if (highmem_pages >= max_pfn) { | |
934 | printk(KERN_ERR "highmem size specified (%uMB) is bigger than pages available (%luMB)!.\n", pages_to_mb(highmem_pages), pages_to_mb(max_pfn)); | |
935 | highmem_pages = 0; | |
936 | } | |
937 | if (highmem_pages) { | |
938 | if (max_low_pfn-highmem_pages < 64*1024*1024/PAGE_SIZE){ | |
939 | printk(KERN_ERR "highmem size %uMB results in smaller than 64MB lowmem, ignoring it.\n", pages_to_mb(highmem_pages)); | |
940 | highmem_pages = 0; | |
941 | } | |
942 | max_low_pfn -= highmem_pages; | |
943 | } | |
944 | #else | |
945 | if (highmem_pages) | |
946 | printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n"); | |
947 | #endif | |
948 | } | |
949 | return max_low_pfn; | |
950 | } | |
951 | ||
952 | /* | |
953 | * Free all available memory for boot time allocation. Used | |
954 | * as a callback function by efi_memory_walk() | |
955 | */ | |
956 | ||
957 | static int __init | |
958 | free_available_memory(unsigned long start, unsigned long end, void *arg) | |
959 | { | |
960 | /* check max_low_pfn */ | |
961 | if (start >= ((max_low_pfn + 1) << PAGE_SHIFT)) | |
962 | return 0; | |
963 | if (end >= ((max_low_pfn + 1) << PAGE_SHIFT)) | |
964 | end = (max_low_pfn + 1) << PAGE_SHIFT; | |
965 | if (start < end) | |
966 | free_bootmem(start, end - start); | |
967 | ||
968 | return 0; | |
969 | } | |
970 | /* | |
971 | * Register fully available low RAM pages with the bootmem allocator. | |
972 | */ | |
973 | static void __init register_bootmem_low_pages(unsigned long max_low_pfn) | |
974 | { | |
975 | int i; | |
976 | ||
977 | if (efi_enabled) { | |
978 | efi_memmap_walk(free_available_memory, NULL); | |
979 | return; | |
980 | } | |
981 | for (i = 0; i < e820.nr_map; i++) { | |
982 | unsigned long curr_pfn, last_pfn, size; | |
983 | /* | |
984 | * Reserve usable low memory | |
985 | */ | |
986 | if (e820.map[i].type != E820_RAM) | |
987 | continue; | |
988 | /* | |
989 | * We are rounding up the start address of usable memory: | |
990 | */ | |
991 | curr_pfn = PFN_UP(e820.map[i].addr); | |
992 | if (curr_pfn >= max_low_pfn) | |
993 | continue; | |
994 | /* | |
995 | * ... and at the end of the usable range downwards: | |
996 | */ | |
997 | last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size); | |
998 | ||
999 | if (last_pfn > max_low_pfn) | |
1000 | last_pfn = max_low_pfn; | |
1001 | ||
1002 | /* | |
1003 | * .. finally, did all the rounding and playing | |
1004 | * around just make the area go away? | |
1005 | */ | |
1006 | if (last_pfn <= curr_pfn) | |
1007 | continue; | |
1008 | ||
1009 | size = last_pfn - curr_pfn; | |
1010 | free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size)); | |
1011 | } | |
1012 | } | |
1013 | ||
1014 | /* | |
1015 | * workaround for Dell systems that neglect to reserve EBDA | |
1016 | */ | |
1017 | static void __init reserve_ebda_region(void) | |
1018 | { | |
1019 | unsigned int addr; | |
1020 | addr = get_bios_ebda(); | |
1021 | if (addr) | |
1022 | reserve_bootmem(addr, PAGE_SIZE); | |
1023 | } | |
1024 | ||
1025 | #ifndef CONFIG_DISCONTIGMEM | |
1026 | void __init setup_bootmem_allocator(void); | |
1027 | static unsigned long __init setup_memory(void) | |
1028 | { | |
1029 | /* | |
1030 | * partially used pages are not usable - thus | |
1031 | * we are rounding upwards: | |
1032 | */ | |
1033 | min_low_pfn = PFN_UP(init_pg_tables_end); | |
1034 | ||
1035 | find_max_pfn(); | |
1036 | ||
1037 | max_low_pfn = find_max_low_pfn(); | |
1038 | ||
1039 | #ifdef CONFIG_HIGHMEM | |
1040 | highstart_pfn = highend_pfn = max_pfn; | |
1041 | if (max_pfn > max_low_pfn) { | |
1042 | highstart_pfn = max_low_pfn; | |
1043 | } | |
1044 | printk(KERN_NOTICE "%ldMB HIGHMEM available.\n", | |
1045 | pages_to_mb(highend_pfn - highstart_pfn)); | |
1046 | #endif | |
1047 | printk(KERN_NOTICE "%ldMB LOWMEM available.\n", | |
1048 | pages_to_mb(max_low_pfn)); | |
1049 | ||
1050 | setup_bootmem_allocator(); | |
1051 | ||
1052 | return max_low_pfn; | |
1053 | } | |
1054 | ||
1055 | void __init zone_sizes_init(void) | |
1056 | { | |
1057 | unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0}; | |
1058 | unsigned int max_dma, low; | |
1059 | ||
1060 | max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT; | |
1061 | low = max_low_pfn; | |
1062 | ||
1063 | if (low < max_dma) | |
1064 | zones_size[ZONE_DMA] = low; | |
1065 | else { | |
1066 | zones_size[ZONE_DMA] = max_dma; | |
1067 | zones_size[ZONE_NORMAL] = low - max_dma; | |
1068 | #ifdef CONFIG_HIGHMEM | |
1069 | zones_size[ZONE_HIGHMEM] = highend_pfn - low; | |
1070 | #endif | |
1071 | } | |
1072 | free_area_init(zones_size); | |
1073 | } | |
1074 | #else | |
1075 | extern unsigned long setup_memory(void); | |
1076 | extern void zone_sizes_init(void); | |
1077 | #endif /* !CONFIG_DISCONTIGMEM */ | |
1078 | ||
1079 | void __init setup_bootmem_allocator(void) | |
1080 | { | |
1081 | unsigned long bootmap_size; | |
1082 | /* | |
1083 | * Initialize the boot-time allocator (with low memory only): | |
1084 | */ | |
1085 | bootmap_size = init_bootmem(min_low_pfn, max_low_pfn); | |
1086 | ||
1087 | register_bootmem_low_pages(max_low_pfn); | |
1088 | ||
1089 | /* | |
1090 | * Reserve the bootmem bitmap itself as well. We do this in two | |
1091 | * steps (first step was init_bootmem()) because this catches | |
1092 | * the (very unlikely) case of us accidentally initializing the | |
1093 | * bootmem allocator with an invalid RAM area. | |
1094 | */ | |
1095 | reserve_bootmem(HIGH_MEMORY, (PFN_PHYS(min_low_pfn) + | |
1096 | bootmap_size + PAGE_SIZE-1) - (HIGH_MEMORY)); | |
1097 | ||
1098 | /* | |
1099 | * reserve physical page 0 - it's a special BIOS page on many boxes, | |
1100 | * enabling clean reboots, SMP operation, laptop functions. | |
1101 | */ | |
1102 | reserve_bootmem(0, PAGE_SIZE); | |
1103 | ||
1104 | /* reserve EBDA region, it's a 4K region */ | |
1105 | reserve_ebda_region(); | |
1106 | ||
1107 | /* could be an AMD 768MPX chipset. Reserve a page before VGA to prevent | |
1108 | PCI prefetch into it (errata #56). Usually the page is reserved anyways, | |
1109 | unless you have no PS/2 mouse plugged in. */ | |
1110 | if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD && | |
1111 | boot_cpu_data.x86 == 6) | |
1112 | reserve_bootmem(0xa0000 - 4096, 4096); | |
1113 | ||
1114 | #ifdef CONFIG_SMP | |
1115 | /* | |
1116 | * But first pinch a few for the stack/trampoline stuff | |
1117 | * FIXME: Don't need the extra page at 4K, but need to fix | |
1118 | * trampoline before removing it. (see the GDT stuff) | |
1119 | */ | |
1120 | reserve_bootmem(PAGE_SIZE, PAGE_SIZE); | |
1121 | #endif | |
1122 | #ifdef CONFIG_ACPI_SLEEP | |
1123 | /* | |
1124 | * Reserve low memory region for sleep support. | |
1125 | */ | |
1126 | acpi_reserve_bootmem(); | |
1127 | #endif | |
1128 | #ifdef CONFIG_X86_FIND_SMP_CONFIG | |
1129 | /* | |
1130 | * Find and reserve possible boot-time SMP configuration: | |
1131 | */ | |
1132 | find_smp_config(); | |
1133 | #endif | |
1134 | ||
1135 | #ifdef CONFIG_BLK_DEV_INITRD | |
1136 | if (LOADER_TYPE && INITRD_START) { | |
1137 | if (INITRD_START + INITRD_SIZE <= (max_low_pfn << PAGE_SHIFT)) { | |
1138 | reserve_bootmem(INITRD_START, INITRD_SIZE); | |
1139 | initrd_start = | |
1140 | INITRD_START ? INITRD_START + PAGE_OFFSET : 0; | |
1141 | initrd_end = initrd_start+INITRD_SIZE; | |
1142 | } | |
1143 | else { | |
1144 | printk(KERN_ERR "initrd extends beyond end of memory " | |
1145 | "(0x%08lx > 0x%08lx)\ndisabling initrd\n", | |
1146 | INITRD_START + INITRD_SIZE, | |
1147 | max_low_pfn << PAGE_SHIFT); | |
1148 | initrd_start = 0; | |
1149 | } | |
1150 | } | |
1151 | #endif | |
1152 | } | |
1153 | ||
1154 | /* | |
1155 | * The node 0 pgdat is initialized before all of these because | |
1156 | * it's needed for bootmem. node>0 pgdats have their virtual | |
1157 | * space allocated before the pagetables are in place to access | |
1158 | * them, so they can't be cleared then. | |
1159 | * | |
1160 | * This should all compile down to nothing when NUMA is off. | |
1161 | */ | |
1162 | void __init remapped_pgdat_init(void) | |
1163 | { | |
1164 | int nid; | |
1165 | ||
1166 | for_each_online_node(nid) { | |
1167 | if (nid != 0) | |
1168 | memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); | |
1169 | } | |
1170 | } | |
1171 | ||
1172 | /* | |
1173 | * Request address space for all standard RAM and ROM resources | |
1174 | * and also for regions reported as reserved by the e820. | |
1175 | */ | |
1176 | static void __init | |
1177 | legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource) | |
1178 | { | |
1179 | int i; | |
1180 | ||
1181 | probe_roms(); | |
1182 | for (i = 0; i < e820.nr_map; i++) { | |
1183 | struct resource *res; | |
1184 | if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL) | |
1185 | continue; | |
1186 | res = alloc_bootmem_low(sizeof(struct resource)); | |
1187 | switch (e820.map[i].type) { | |
1188 | case E820_RAM: res->name = "System RAM"; break; | |
1189 | case E820_ACPI: res->name = "ACPI Tables"; break; | |
1190 | case E820_NVS: res->name = "ACPI Non-volatile Storage"; break; | |
1191 | default: res->name = "reserved"; | |
1192 | } | |
1193 | res->start = e820.map[i].addr; | |
1194 | res->end = res->start + e820.map[i].size - 1; | |
1195 | res->flags = IORESOURCE_MEM | IORESOURCE_BUSY; | |
1196 | request_resource(&iomem_resource, res); | |
1197 | if (e820.map[i].type == E820_RAM) { | |
1198 | /* | |
1199 | * We don't know which RAM region contains kernel data, | |
1200 | * so we try it repeatedly and let the resource manager | |
1201 | * test it. | |
1202 | */ | |
1203 | request_resource(res, code_resource); | |
1204 | request_resource(res, data_resource); | |
1205 | } | |
1206 | } | |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * Request address space for all standard resources | |
1211 | */ | |
1212 | static void __init register_memory(void) | |
1213 | { | |
1214 | unsigned long gapstart, gapsize; | |
1215 | unsigned long long last; | |
1216 | int i; | |
1217 | ||
1218 | if (efi_enabled) | |
1219 | efi_initialize_iomem_resources(&code_resource, &data_resource); | |
1220 | else | |
1221 | legacy_init_iomem_resources(&code_resource, &data_resource); | |
1222 | ||
1223 | /* EFI systems may still have VGA */ | |
1224 | request_resource(&iomem_resource, &video_ram_resource); | |
1225 | ||
1226 | /* request I/O space for devices used on all i[345]86 PCs */ | |
1227 | for (i = 0; i < STANDARD_IO_RESOURCES; i++) | |
1228 | request_resource(&ioport_resource, &standard_io_resources[i]); | |
1229 | ||
1230 | /* | |
1231 | * Search for the bigest gap in the low 32 bits of the e820 | |
1232 | * memory space. | |
1233 | */ | |
1234 | last = 0x100000000ull; | |
1235 | gapstart = 0x10000000; | |
1236 | gapsize = 0x400000; | |
1237 | i = e820.nr_map; | |
1238 | while (--i >= 0) { | |
1239 | unsigned long long start = e820.map[i].addr; | |
1240 | unsigned long long end = start + e820.map[i].size; | |
1241 | ||
1242 | /* | |
1243 | * Since "last" is at most 4GB, we know we'll | |
1244 | * fit in 32 bits if this condition is true | |
1245 | */ | |
1246 | if (last > end) { | |
1247 | unsigned long gap = last - end; | |
1248 | ||
1249 | if (gap > gapsize) { | |
1250 | gapsize = gap; | |
1251 | gapstart = end; | |
1252 | } | |
1253 | } | |
1254 | if (start < last) | |
1255 | last = start; | |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * Start allocating dynamic PCI memory a bit into the gap, | |
1260 | * aligned up to the nearest megabyte. | |
1261 | * | |
1262 | * Question: should we try to pad it up a bit (do something | |
1263 | * like " + (gapsize >> 3)" in there too?). We now have the | |
1264 | * technology. | |
1265 | */ | |
1266 | pci_mem_start = (gapstart + 0xfffff) & ~0xfffff; | |
1267 | ||
1268 | printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n", | |
1269 | pci_mem_start, gapstart, gapsize); | |
1270 | } | |
1271 | ||
1272 | /* Use inline assembly to define this because the nops are defined | |
1273 | as inline assembly strings in the include files and we cannot | |
1274 | get them easily into strings. */ | |
1275 | asm("\t.data\nintelnops: " | |
1276 | GENERIC_NOP1 GENERIC_NOP2 GENERIC_NOP3 GENERIC_NOP4 GENERIC_NOP5 GENERIC_NOP6 | |
1277 | GENERIC_NOP7 GENERIC_NOP8); | |
1278 | asm("\t.data\nk8nops: " | |
1279 | K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6 | |
1280 | K8_NOP7 K8_NOP8); | |
1281 | asm("\t.data\nk7nops: " | |
1282 | K7_NOP1 K7_NOP2 K7_NOP3 K7_NOP4 K7_NOP5 K7_NOP6 | |
1283 | K7_NOP7 K7_NOP8); | |
1284 | ||
1285 | extern unsigned char intelnops[], k8nops[], k7nops[]; | |
1286 | static unsigned char *intel_nops[ASM_NOP_MAX+1] = { | |
1287 | NULL, | |
1288 | intelnops, | |
1289 | intelnops + 1, | |
1290 | intelnops + 1 + 2, | |
1291 | intelnops + 1 + 2 + 3, | |
1292 | intelnops + 1 + 2 + 3 + 4, | |
1293 | intelnops + 1 + 2 + 3 + 4 + 5, | |
1294 | intelnops + 1 + 2 + 3 + 4 + 5 + 6, | |
1295 | intelnops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
1296 | }; | |
1297 | static unsigned char *k8_nops[ASM_NOP_MAX+1] = { | |
1298 | NULL, | |
1299 | k8nops, | |
1300 | k8nops + 1, | |
1301 | k8nops + 1 + 2, | |
1302 | k8nops + 1 + 2 + 3, | |
1303 | k8nops + 1 + 2 + 3 + 4, | |
1304 | k8nops + 1 + 2 + 3 + 4 + 5, | |
1305 | k8nops + 1 + 2 + 3 + 4 + 5 + 6, | |
1306 | k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
1307 | }; | |
1308 | static unsigned char *k7_nops[ASM_NOP_MAX+1] = { | |
1309 | NULL, | |
1310 | k7nops, | |
1311 | k7nops + 1, | |
1312 | k7nops + 1 + 2, | |
1313 | k7nops + 1 + 2 + 3, | |
1314 | k7nops + 1 + 2 + 3 + 4, | |
1315 | k7nops + 1 + 2 + 3 + 4 + 5, | |
1316 | k7nops + 1 + 2 + 3 + 4 + 5 + 6, | |
1317 | k7nops + 1 + 2 + 3 + 4 + 5 + 6 + 7, | |
1318 | }; | |
1319 | static struct nop { | |
1320 | int cpuid; | |
1321 | unsigned char **noptable; | |
1322 | } noptypes[] = { | |
1323 | { X86_FEATURE_K8, k8_nops }, | |
1324 | { X86_FEATURE_K7, k7_nops }, | |
1325 | { -1, NULL } | |
1326 | }; | |
1327 | ||
1328 | /* Replace instructions with better alternatives for this CPU type. | |
1329 | ||
1330 | This runs before SMP is initialized to avoid SMP problems with | |
1331 | self modifying code. This implies that assymetric systems where | |
1332 | APs have less capabilities than the boot processor are not handled. | |
1333 | In this case boot with "noreplacement". */ | |
1334 | void apply_alternatives(void *start, void *end) | |
1335 | { | |
1336 | struct alt_instr *a; | |
1337 | int diff, i, k; | |
1338 | unsigned char **noptable = intel_nops; | |
1339 | for (i = 0; noptypes[i].cpuid >= 0; i++) { | |
1340 | if (boot_cpu_has(noptypes[i].cpuid)) { | |
1341 | noptable = noptypes[i].noptable; | |
1342 | break; | |
1343 | } | |
1344 | } | |
1345 | for (a = start; (void *)a < end; a++) { | |
1346 | if (!boot_cpu_has(a->cpuid)) | |
1347 | continue; | |
1348 | BUG_ON(a->replacementlen > a->instrlen); | |
1349 | memcpy(a->instr, a->replacement, a->replacementlen); | |
1350 | diff = a->instrlen - a->replacementlen; | |
1351 | /* Pad the rest with nops */ | |
1352 | for (i = a->replacementlen; diff > 0; diff -= k, i += k) { | |
1353 | k = diff; | |
1354 | if (k > ASM_NOP_MAX) | |
1355 | k = ASM_NOP_MAX; | |
1356 | memcpy(a->instr + i, noptable[k], k); | |
1357 | } | |
1358 | } | |
1359 | } | |
1360 | ||
1361 | static int no_replacement __initdata = 0; | |
1362 | ||
1363 | void __init alternative_instructions(void) | |
1364 | { | |
1365 | extern struct alt_instr __alt_instructions[], __alt_instructions_end[]; | |
1366 | if (no_replacement) | |
1367 | return; | |
1368 | apply_alternatives(__alt_instructions, __alt_instructions_end); | |
1369 | } | |
1370 | ||
1371 | static int __init noreplacement_setup(char *s) | |
1372 | { | |
1373 | no_replacement = 1; | |
1374 | return 0; | |
1375 | } | |
1376 | ||
1377 | __setup("noreplacement", noreplacement_setup); | |
1378 | ||
1379 | static char * __init machine_specific_memory_setup(void); | |
1380 | ||
1381 | #ifdef CONFIG_MCA | |
1382 | static void set_mca_bus(int x) | |
1383 | { | |
1384 | MCA_bus = x; | |
1385 | } | |
1386 | #else | |
1387 | static void set_mca_bus(int x) { } | |
1388 | #endif | |
1389 | ||
1390 | /* | |
1391 | * Determine if we were loaded by an EFI loader. If so, then we have also been | |
1392 | * passed the efi memmap, systab, etc., so we should use these data structures | |
1393 | * for initialization. Note, the efi init code path is determined by the | |
1394 | * global efi_enabled. This allows the same kernel image to be used on existing | |
1395 | * systems (with a traditional BIOS) as well as on EFI systems. | |
1396 | */ | |
1397 | void __init setup_arch(char **cmdline_p) | |
1398 | { | |
1399 | unsigned long max_low_pfn; | |
1400 | ||
1401 | memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data)); | |
1402 | pre_setup_arch_hook(); | |
1403 | early_cpu_init(); | |
1404 | ||
1405 | /* | |
1406 | * FIXME: This isn't an official loader_type right | |
1407 | * now but does currently work with elilo. | |
1408 | * If we were configured as an EFI kernel, check to make | |
1409 | * sure that we were loaded correctly from elilo and that | |
1410 | * the system table is valid. If not, then initialize normally. | |
1411 | */ | |
1412 | #ifdef CONFIG_EFI | |
1413 | if ((LOADER_TYPE == 0x50) && EFI_SYSTAB) | |
1414 | efi_enabled = 1; | |
1415 | #endif | |
1416 | ||
1417 | ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV); | |
1418 | drive_info = DRIVE_INFO; | |
1419 | screen_info = SCREEN_INFO; | |
1420 | edid_info = EDID_INFO; | |
1421 | apm_info.bios = APM_BIOS_INFO; | |
1422 | ist_info = IST_INFO; | |
1423 | saved_videomode = VIDEO_MODE; | |
1424 | if( SYS_DESC_TABLE.length != 0 ) { | |
1425 | set_mca_bus(SYS_DESC_TABLE.table[3] & 0x2); | |
1426 | machine_id = SYS_DESC_TABLE.table[0]; | |
1427 | machine_submodel_id = SYS_DESC_TABLE.table[1]; | |
1428 | BIOS_revision = SYS_DESC_TABLE.table[2]; | |
1429 | } | |
1430 | bootloader_type = LOADER_TYPE; | |
1431 | ||
1432 | #ifdef CONFIG_BLK_DEV_RAM | |
1433 | rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK; | |
1434 | rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0); | |
1435 | rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0); | |
1436 | #endif | |
1437 | ARCH_SETUP | |
1438 | if (efi_enabled) | |
1439 | efi_init(); | |
1440 | else { | |
1441 | printk(KERN_INFO "BIOS-provided physical RAM map:\n"); | |
1442 | print_memory_map(machine_specific_memory_setup()); | |
1443 | } | |
1444 | ||
1445 | copy_edd(); | |
1446 | ||
1447 | if (!MOUNT_ROOT_RDONLY) | |
1448 | root_mountflags &= ~MS_RDONLY; | |
1449 | init_mm.start_code = (unsigned long) _text; | |
1450 | init_mm.end_code = (unsigned long) _etext; | |
1451 | init_mm.end_data = (unsigned long) _edata; | |
1452 | init_mm.brk = init_pg_tables_end + PAGE_OFFSET; | |
1453 | ||
1454 | code_resource.start = virt_to_phys(_text); | |
1455 | code_resource.end = virt_to_phys(_etext)-1; | |
1456 | data_resource.start = virt_to_phys(_etext); | |
1457 | data_resource.end = virt_to_phys(_edata)-1; | |
1458 | ||
1459 | parse_cmdline_early(cmdline_p); | |
1460 | ||
1461 | max_low_pfn = setup_memory(); | |
1462 | ||
1463 | /* | |
1464 | * NOTE: before this point _nobody_ is allowed to allocate | |
1465 | * any memory using the bootmem allocator. Although the | |
1466 | * alloctor is now initialised only the first 8Mb of the kernel | |
1467 | * virtual address space has been mapped. All allocations before | |
1468 | * paging_init() has completed must use the alloc_bootmem_low_pages() | |
1469 | * variant (which allocates DMA'able memory) and care must be taken | |
1470 | * not to exceed the 8Mb limit. | |
1471 | */ | |
1472 | ||
1473 | #ifdef CONFIG_SMP | |
1474 | smp_alloc_memory(); /* AP processor realmode stacks in low memory*/ | |
1475 | #endif | |
1476 | paging_init(); | |
1477 | remapped_pgdat_init(); | |
1478 | zone_sizes_init(); | |
1479 | ||
1480 | /* | |
1481 | * NOTE: at this point the bootmem allocator is fully available. | |
1482 | */ | |
1483 | ||
1484 | #ifdef CONFIG_EARLY_PRINTK | |
1485 | { | |
1486 | char *s = strstr(*cmdline_p, "earlyprintk="); | |
1487 | if (s) { | |
1488 | extern void setup_early_printk(char *); | |
1489 | ||
1490 | setup_early_printk(s); | |
1491 | printk("early console enabled\n"); | |
1492 | } | |
1493 | } | |
1494 | #endif | |
1495 | ||
1496 | ||
1497 | dmi_scan_machine(); | |
1498 | ||
1499 | #ifdef CONFIG_X86_GENERICARCH | |
1500 | generic_apic_probe(*cmdline_p); | |
1501 | #endif | |
1502 | if (efi_enabled) | |
1503 | efi_map_memmap(); | |
1504 | ||
1505 | /* | |
1506 | * Parse the ACPI tables for possible boot-time SMP configuration. | |
1507 | */ | |
1508 | acpi_boot_table_init(); | |
1509 | acpi_boot_init(); | |
1510 | ||
1511 | #ifdef CONFIG_X86_LOCAL_APIC | |
1512 | if (smp_found_config) | |
1513 | get_smp_config(); | |
1514 | #endif | |
1515 | ||
1516 | register_memory(); | |
1517 | ||
1518 | #ifdef CONFIG_VT | |
1519 | #if defined(CONFIG_VGA_CONSOLE) | |
1520 | if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY)) | |
1521 | conswitchp = &vga_con; | |
1522 | #elif defined(CONFIG_DUMMY_CONSOLE) | |
1523 | conswitchp = &dummy_con; | |
1524 | #endif | |
1525 | #endif | |
1526 | } | |
1527 | ||
1528 | #include "setup_arch_post.h" | |
1529 | /* | |
1530 | * Local Variables: | |
1531 | * mode:c | |
1532 | * c-file-style:"k&r" | |
1533 | * c-basic-offset:8 | |
1534 | * End: | |
1535 | */ |