1 /*
2 * Copyright (C) 1995 Linus Torvalds
3 *
4 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
5 *
6 * Memory region support
7 * David Parsons <orc@pell.chi.il.us>, July-August 1999
8 *
9 * Added E820 sanitization routine (removes overlapping memory regions);
10 * Brian Moyle <bmoyle@mvista.com>, February 2001
11 *
12 * Moved CPU detection code to cpu/${cpu}.c
13 * Patrick Mochel <mochel@osdl.org>, March 2002
14 *
15 * Provisions for empty E820 memory regions (reported by certain BIOSes).
16 * Alex Achenbach <xela@slit.de>, December 2002.
17 *
18 */
19
20 /*
21 * This file handles the architecture-dependent parts of initialization
22 */
23
24 #include <linux/sched.h>
25 #include <linux/mm.h>
26 #include <linux/mmzone.h>
27 #include <linux/screen_info.h>
28 #include <linux/ioport.h>
29 #include <linux/acpi.h>
30 #include <linux/sfi.h>
31 #include <linux/apm_bios.h>
32 #include <linux/initrd.h>
33 #include <linux/bootmem.h>
34 #include <linux/memblock.h>
35 #include <linux/seq_file.h>
36 #include <linux/console.h>
37 #include <linux/mca.h>
38 #include <linux/root_dev.h>
39 #include <linux/highmem.h>
40 #include <linux/module.h>
41 #include <linux/efi.h>
42 #include <linux/init.h>
43 #include <linux/edd.h>
44 #include <linux/iscsi_ibft.h>
45 #include <linux/nodemask.h>
46 #include <linux/kexec.h>
47 #include <linux/dmi.h>
48 #include <linux/pfn.h>
49 #include <linux/pci.h>
50 #include <asm/pci-direct.h>
51 #include <linux/init_ohci1394_dma.h>
52 #include <linux/kvm_para.h>
53
54 #include <linux/errno.h>
55 #include <linux/kernel.h>
56 #include <linux/stddef.h>
57 #include <linux/unistd.h>
58 #include <linux/ptrace.h>
59 #include <linux/user.h>
60 #include <linux/delay.h>
61
62 #include <linux/kallsyms.h>
63 #include <linux/cpufreq.h>
64 #include <linux/dma-mapping.h>
65 #include <linux/ctype.h>
66 #include <linux/uaccess.h>
67
68 #include <linux/percpu.h>
69 #include <linux/crash_dump.h>
70 #include <linux/tboot.h>
71
72 #include <video/edid.h>
73
74 #include <asm/mtrr.h>
75 #include <asm/apic.h>
76 #include <asm/trampoline.h>
77 #include <asm/e820.h>
78 #include <asm/mpspec.h>
79 #include <asm/setup.h>
80 #include <asm/efi.h>
81 #include <asm/timer.h>
82 #include <asm/i8259.h>
83 #include <asm/sections.h>
84 #include <asm/dmi.h>
85 #include <asm/io_apic.h>
86 #include <asm/ist.h>
87 #include <asm/setup_arch.h>
88 #include <asm/bios_ebda.h>
89 #include <asm/cacheflush.h>
90 #include <asm/processor.h>
91 #include <asm/bugs.h>
92
93 #include <asm/vsyscall.h>
94 #include <asm/cpu.h>
95 #include <asm/desc.h>
96 #include <asm/dma.h>
97 #include <asm/iommu.h>
98 #include <asm/gart.h>
99 #include <asm/mmu_context.h>
100 #include <asm/proto.h>
101
102 #include <asm/paravirt.h>
103 #include <asm/hypervisor.h>
104 #include <asm/olpc_ofw.h>
105
106 #include <asm/percpu.h>
107 #include <asm/topology.h>
108 #include <asm/apicdef.h>
109 #include <asm/amd_nb.h>
110 #ifdef CONFIG_X86_64
111 #include <asm/numa_64.h>
112 #endif
113 #include <asm/mce.h>
114 #include <asm/alternative.h>
115 #include <asm/prom.h>
116
117 /*
118 * end_pfn only includes RAM, while max_pfn_mapped includes all e820 entries.
119 * The direct mapping extends to max_pfn_mapped, so that we can directly access
120 * apertures, ACPI and other tables without having to play with fixmaps.
121 */
122 unsigned long max_low_pfn_mapped;
123 unsigned long max_pfn_mapped;
124
125 #ifdef CONFIG_DMI
126 RESERVE_BRK(dmi_alloc, 65536);
127 #endif
128
129
130 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
131 unsigned long _brk_end = (unsigned long)__brk_base;
132
133 #ifdef CONFIG_X86_64
default_cpu_present_to_apicid(int mps_cpu)134 int default_cpu_present_to_apicid(int mps_cpu)
135 {
136 return __default_cpu_present_to_apicid(mps_cpu);
137 }
138
default_check_phys_apicid_present(int phys_apicid)139 int default_check_phys_apicid_present(int phys_apicid)
140 {
141 return __default_check_phys_apicid_present(phys_apicid);
142 }
143 #endif
144
145 #ifndef CONFIG_DEBUG_BOOT_PARAMS
146 struct boot_params __initdata boot_params;
147 #else
148 struct boot_params boot_params;
149 #endif
150
151 /*
152 * Machine setup..
153 */
154 static struct resource data_resource = {
155 .name = "Kernel data",
156 .start = 0,
157 .end = 0,
158 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
159 };
160
161 static struct resource code_resource = {
162 .name = "Kernel code",
163 .start = 0,
164 .end = 0,
165 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
166 };
167
168 static struct resource bss_resource = {
169 .name = "Kernel bss",
170 .start = 0,
171 .end = 0,
172 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
173 };
174
175
176 #ifdef CONFIG_X86_32
177 /* cpu data as detected by the assembly code in head.S */
178 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {0, 0, 0, 0, -1, 1, 0, 0, -1};
179 /* common cpu data for all cpus */
180 struct cpuinfo_x86 boot_cpu_data __read_mostly = {0, 0, 0, 0, -1, 1, 0, 0, -1};
181 EXPORT_SYMBOL(boot_cpu_data);
set_mca_bus(int x)182 static void set_mca_bus(int x)
183 {
184 #ifdef CONFIG_MCA
185 MCA_bus = x;
186 #endif
187 }
188
189 unsigned int def_to_bigsmp;
190
191 /* for MCA, but anyone else can use it if they want */
192 unsigned int machine_id;
193 unsigned int machine_submodel_id;
194 unsigned int BIOS_revision;
195
196 struct apm_info apm_info;
197 EXPORT_SYMBOL(apm_info);
198
199 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
200 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
201 struct ist_info ist_info;
202 EXPORT_SYMBOL(ist_info);
203 #else
204 struct ist_info ist_info;
205 #endif
206
207 #else
208 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
209 .x86_phys_bits = MAX_PHYSMEM_BITS,
210 };
211 EXPORT_SYMBOL(boot_cpu_data);
212 #endif
213
214
215 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
216 unsigned long mmu_cr4_features;
217 #else
218 unsigned long mmu_cr4_features = X86_CR4_PAE;
219 #endif
220
221 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
222 int bootloader_type, bootloader_version;
223
224 /*
225 * Setup options
226 */
227 struct screen_info screen_info;
228 EXPORT_SYMBOL(screen_info);
229 struct edid_info edid_info;
230 EXPORT_SYMBOL_GPL(edid_info);
231
232 extern int root_mountflags;
233
234 unsigned long saved_video_mode;
235
236 #define RAMDISK_IMAGE_START_MASK 0x07FF
237 #define RAMDISK_PROMPT_FLAG 0x8000
238 #define RAMDISK_LOAD_FLAG 0x4000
239
240 static char __initdata command_line[COMMAND_LINE_SIZE];
241 #ifdef CONFIG_CMDLINE_BOOL
242 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
243 #endif
244
245 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
246 struct edd edd;
247 #ifdef CONFIG_EDD_MODULE
248 EXPORT_SYMBOL(edd);
249 #endif
250 /**
251 * copy_edd() - Copy the BIOS EDD information
252 * from boot_params into a safe place.
253 *
254 */
copy_edd(void)255 static inline void __init copy_edd(void)
256 {
257 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
258 sizeof(edd.mbr_signature));
259 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
260 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
261 edd.edd_info_nr = boot_params.eddbuf_entries;
262 }
263 #else
copy_edd(void)264 static inline void __init copy_edd(void)
265 {
266 }
267 #endif
268
extend_brk(size_t size,size_t align)269 void * __init extend_brk(size_t size, size_t align)
270 {
271 size_t mask = align - 1;
272 void *ret;
273
274 BUG_ON(_brk_start == 0);
275 BUG_ON(align & mask);
276
277 _brk_end = (_brk_end + mask) & ~mask;
278 BUG_ON((char *)(_brk_end + size) > __brk_limit);
279
280 ret = (void *)_brk_end;
281 _brk_end += size;
282
283 memset(ret, 0, size);
284
285 return ret;
286 }
287
288 #ifdef CONFIG_X86_64
init_gbpages(void)289 static void __init init_gbpages(void)
290 {
291 if (direct_gbpages && cpu_has_gbpages)
292 printk(KERN_INFO "Using GB pages for direct mapping\n");
293 else
294 direct_gbpages = 0;
295 }
296 #else
init_gbpages(void)297 static inline void init_gbpages(void)
298 {
299 }
cleanup_highmap(void)300 static void __init cleanup_highmap(void)
301 {
302 }
303 #endif
304
reserve_brk(void)305 static void __init reserve_brk(void)
306 {
307 if (_brk_end > _brk_start)
308 memblock_reserve(__pa(_brk_start),
309 __pa(_brk_end) - __pa(_brk_start));
310
311 /* Mark brk area as locked down and no longer taking any
312 new allocations */
313 _brk_start = 0;
314 }
315
316 #ifdef CONFIG_BLK_DEV_INITRD
317
318 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
relocate_initrd(void)319 static void __init relocate_initrd(void)
320 {
321 /* Assume only end is not page aligned */
322 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
323 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
324 u64 area_size = PAGE_ALIGN(ramdisk_size);
325 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
326 u64 ramdisk_here;
327 unsigned long slop, clen, mapaddr;
328 char *p, *q;
329
330 /* We need to move the initrd down into lowmem */
331 ramdisk_here = memblock_find_in_range(0, end_of_lowmem, area_size,
332 PAGE_SIZE);
333
334 if (!ramdisk_here)
335 panic("Cannot find place for new RAMDISK of size %lld\n",
336 ramdisk_size);
337
338 /* Note: this includes all the lowmem currently occupied by
339 the initrd, we rely on that fact to keep the data intact. */
340 memblock_reserve(ramdisk_here, area_size);
341 initrd_start = ramdisk_here + PAGE_OFFSET;
342 initrd_end = initrd_start + ramdisk_size;
343 printk(KERN_INFO "Allocated new RAMDISK: %08llx - %08llx\n",
344 ramdisk_here, ramdisk_here + ramdisk_size);
345
346 q = (char *)initrd_start;
347
348 /* Copy any lowmem portion of the initrd */
349 if (ramdisk_image < end_of_lowmem) {
350 clen = end_of_lowmem - ramdisk_image;
351 p = (char *)__va(ramdisk_image);
352 memcpy(q, p, clen);
353 q += clen;
354 ramdisk_image += clen;
355 ramdisk_size -= clen;
356 }
357
358 /* Copy the highmem portion of the initrd */
359 while (ramdisk_size) {
360 slop = ramdisk_image & ~PAGE_MASK;
361 clen = ramdisk_size;
362 if (clen > MAX_MAP_CHUNK-slop)
363 clen = MAX_MAP_CHUNK-slop;
364 mapaddr = ramdisk_image & PAGE_MASK;
365 p = early_memremap(mapaddr, clen+slop);
366 memcpy(q, p+slop, clen);
367 early_iounmap(p, clen+slop);
368 q += clen;
369 ramdisk_image += clen;
370 ramdisk_size -= clen;
371 }
372 /* high pages is not converted by early_res_to_bootmem */
373 ramdisk_image = boot_params.hdr.ramdisk_image;
374 ramdisk_size = boot_params.hdr.ramdisk_size;
375 printk(KERN_INFO "Move RAMDISK from %016llx - %016llx to"
376 " %08llx - %08llx\n",
377 ramdisk_image, ramdisk_image + ramdisk_size - 1,
378 ramdisk_here, ramdisk_here + ramdisk_size - 1);
379 }
380
reserve_initrd(void)381 static void __init reserve_initrd(void)
382 {
383 /* Assume only end is not page aligned */
384 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
385 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
386 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
387 u64 end_of_lowmem = max_low_pfn_mapped << PAGE_SHIFT;
388
389 if (!boot_params.hdr.type_of_loader ||
390 !ramdisk_image || !ramdisk_size)
391 return; /* No initrd provided by bootloader */
392
393 initrd_start = 0;
394
395 if (ramdisk_size >= (end_of_lowmem>>1)) {
396 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
397 printk(KERN_ERR "initrd too large to handle, "
398 "disabling initrd\n");
399 return;
400 }
401
402 printk(KERN_INFO "RAMDISK: %08llx - %08llx\n", ramdisk_image,
403 ramdisk_end);
404
405
406 if (ramdisk_end <= end_of_lowmem) {
407 /* All in lowmem, easy case */
408 /*
409 * don't need to reserve again, already reserved early
410 * in i386_start_kernel
411 */
412 initrd_start = ramdisk_image + PAGE_OFFSET;
413 initrd_end = initrd_start + ramdisk_size;
414 return;
415 }
416
417 relocate_initrd();
418
419 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
420 }
421 #else
reserve_initrd(void)422 static void __init reserve_initrd(void)
423 {
424 }
425 #endif /* CONFIG_BLK_DEV_INITRD */
426
parse_setup_data(void)427 static void __init parse_setup_data(void)
428 {
429 struct setup_data *data;
430 u64 pa_data;
431
432 if (boot_params.hdr.version < 0x0209)
433 return;
434 pa_data = boot_params.hdr.setup_data;
435 while (pa_data) {
436 u32 data_len, map_len;
437
438 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
439 (u64)sizeof(struct setup_data));
440 data = early_memremap(pa_data, map_len);
441 data_len = data->len + sizeof(struct setup_data);
442 if (data_len > map_len) {
443 early_iounmap(data, map_len);
444 data = early_memremap(pa_data, data_len);
445 map_len = data_len;
446 }
447
448 switch (data->type) {
449 case SETUP_E820_EXT:
450 parse_e820_ext(data);
451 break;
452 case SETUP_DTB:
453 add_dtb(pa_data);
454 break;
455 default:
456 break;
457 }
458 pa_data = data->next;
459 early_iounmap(data, map_len);
460 }
461 }
462
e820_reserve_setup_data(void)463 static void __init e820_reserve_setup_data(void)
464 {
465 struct setup_data *data;
466 u64 pa_data;
467 int found = 0;
468
469 if (boot_params.hdr.version < 0x0209)
470 return;
471 pa_data = boot_params.hdr.setup_data;
472 while (pa_data) {
473 data = early_memremap(pa_data, sizeof(*data));
474 e820_update_range(pa_data, sizeof(*data)+data->len,
475 E820_RAM, E820_RESERVED_KERN);
476 found = 1;
477 pa_data = data->next;
478 early_iounmap(data, sizeof(*data));
479 }
480 if (!found)
481 return;
482
483 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
484 memcpy(&e820_saved, &e820, sizeof(struct e820map));
485 printk(KERN_INFO "extended physical RAM map:\n");
486 e820_print_map("reserve setup_data");
487 }
488
memblock_x86_reserve_range_setup_data(void)489 static void __init memblock_x86_reserve_range_setup_data(void)
490 {
491 struct setup_data *data;
492 u64 pa_data;
493
494 if (boot_params.hdr.version < 0x0209)
495 return;
496 pa_data = boot_params.hdr.setup_data;
497 while (pa_data) {
498 data = early_memremap(pa_data, sizeof(*data));
499 memblock_reserve(pa_data, sizeof(*data) + data->len);
500 pa_data = data->next;
501 early_iounmap(data, sizeof(*data));
502 }
503 }
504
505 /*
506 * --------- Crashkernel reservation ------------------------------
507 */
508
509 #ifdef CONFIG_KEXEC
510
511 /*
512 * Keep the crash kernel below this limit. On 32 bits earlier kernels
513 * would limit the kernel to the low 512 MiB due to mapping restrictions.
514 * On 64 bits, kexec-tools currently limits us to 896 MiB; increase this
515 * limit once kexec-tools are fixed.
516 */
517 #ifdef CONFIG_X86_32
518 # define CRASH_KERNEL_ADDR_MAX (512 << 20)
519 #else
520 # define CRASH_KERNEL_ADDR_MAX (896 << 20)
521 #endif
522
reserve_crashkernel(void)523 static void __init reserve_crashkernel(void)
524 {
525 unsigned long long total_mem;
526 unsigned long long crash_size, crash_base;
527 int ret;
528
529 total_mem = memblock_phys_mem_size();
530
531 ret = parse_crashkernel(boot_command_line, total_mem,
532 &crash_size, &crash_base);
533 if (ret != 0 || crash_size <= 0)
534 return;
535
536 /* 0 means: find the address automatically */
537 if (crash_base <= 0) {
538 const unsigned long long alignment = 16<<20; /* 16M */
539
540 /*
541 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
542 */
543 crash_base = memblock_find_in_range(alignment,
544 CRASH_KERNEL_ADDR_MAX, crash_size, alignment);
545
546 if (!crash_base) {
547 pr_info("crashkernel reservation failed - No suitable area found.\n");
548 return;
549 }
550 } else {
551 unsigned long long start;
552
553 start = memblock_find_in_range(crash_base,
554 crash_base + crash_size, crash_size, 1<<20);
555 if (start != crash_base) {
556 pr_info("crashkernel reservation failed - memory is in use.\n");
557 return;
558 }
559 }
560 memblock_reserve(crash_base, crash_size);
561
562 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
563 "for crashkernel (System RAM: %ldMB)\n",
564 (unsigned long)(crash_size >> 20),
565 (unsigned long)(crash_base >> 20),
566 (unsigned long)(total_mem >> 20));
567
568 crashk_res.start = crash_base;
569 crashk_res.end = crash_base + crash_size - 1;
570 insert_resource(&iomem_resource, &crashk_res);
571 }
572 #else
reserve_crashkernel(void)573 static void __init reserve_crashkernel(void)
574 {
575 }
576 #endif
577
578 static struct resource standard_io_resources[] = {
579 { .name = "dma1", .start = 0x00, .end = 0x1f,
580 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
581 { .name = "pic1", .start = 0x20, .end = 0x21,
582 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
583 { .name = "timer0", .start = 0x40, .end = 0x43,
584 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
585 { .name = "timer1", .start = 0x50, .end = 0x53,
586 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
587 { .name = "keyboard", .start = 0x60, .end = 0x60,
588 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
589 { .name = "keyboard", .start = 0x64, .end = 0x64,
590 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
591 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
592 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
593 { .name = "pic2", .start = 0xa0, .end = 0xa1,
594 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
595 { .name = "dma2", .start = 0xc0, .end = 0xdf,
596 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
597 { .name = "fpu", .start = 0xf0, .end = 0xff,
598 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
599 };
600
reserve_standard_io_resources(void)601 void __init reserve_standard_io_resources(void)
602 {
603 int i;
604
605 /* request I/O space for devices used on all i[345]86 PCs */
606 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
607 request_resource(&ioport_resource, &standard_io_resources[i]);
608
609 }
610
reserve_ibft_region(void)611 static __init void reserve_ibft_region(void)
612 {
613 unsigned long addr, size = 0;
614
615 addr = find_ibft_region(&size);
616
617 if (size)
618 memblock_reserve(addr, size);
619 }
620
621 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
622
snb_gfx_workaround_needed(void)623 static bool __init snb_gfx_workaround_needed(void)
624 {
625 #ifdef CONFIG_PCI
626 int i;
627 u16 vendor, devid;
628 static const u16 snb_ids[] = {
629 0x0102,
630 0x0112,
631 0x0122,
632 0x0106,
633 0x0116,
634 0x0126,
635 0x010a,
636 };
637
638 /* Assume no if something weird is going on with PCI */
639 if (!early_pci_allowed())
640 return false;
641
642 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
643 if (vendor != 0x8086)
644 return false;
645
646 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
647 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
648 if (devid == snb_ids[i])
649 return true;
650 #endif
651
652 return false;
653 }
654
655 /*
656 * Sandy Bridge graphics has trouble with certain ranges, exclude
657 * them from allocation.
658 */
trim_snb_memory(void)659 static void __init trim_snb_memory(void)
660 {
661 static const unsigned long bad_pages[] = {
662 0x20050000,
663 0x20110000,
664 0x20130000,
665 0x20138000,
666 0x40004000,
667 };
668 int i;
669
670 if (!snb_gfx_workaround_needed())
671 return;
672
673 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
674
675 /*
676 * Reserve all memory below the 1 MB mark that has not
677 * already been reserved.
678 */
679 memblock_reserve(0, 1<<20);
680
681 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
682 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
683 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
684 bad_pages[i]);
685 }
686 }
687
688 /*
689 * Here we put platform-specific memory range workarounds, i.e.
690 * memory known to be corrupt or otherwise in need to be reserved on
691 * specific platforms.
692 *
693 * If this gets used more widely it could use a real dispatch mechanism.
694 */
trim_platform_memory_ranges(void)695 static void __init trim_platform_memory_ranges(void)
696 {
697 trim_snb_memory();
698 }
699
trim_bios_range(void)700 static void __init trim_bios_range(void)
701 {
702 /*
703 * A special case is the first 4Kb of memory;
704 * This is a BIOS owned area, not kernel ram, but generally
705 * not listed as such in the E820 table.
706 *
707 * This typically reserves additional memory (64KiB by default)
708 * since some BIOSes are known to corrupt low memory. See the
709 * Kconfig help text for X86_RESERVE_LOW.
710 */
711 e820_update_range(0, ALIGN(reserve_low, PAGE_SIZE),
712 E820_RAM, E820_RESERVED);
713
714 /*
715 * special case: Some BIOSen report the PC BIOS
716 * area (640->1Mb) as ram even though it is not.
717 * take them out.
718 */
719 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
720
721 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
722 }
723
parse_reservelow(char * p)724 static int __init parse_reservelow(char *p)
725 {
726 unsigned long long size;
727
728 if (!p)
729 return -EINVAL;
730
731 size = memparse(p, &p);
732
733 if (size < 4096)
734 size = 4096;
735
736 if (size > 640*1024)
737 size = 640*1024;
738
739 reserve_low = size;
740
741 return 0;
742 }
743
744 early_param("reservelow", parse_reservelow);
745
746 /*
747 * Determine if we were loaded by an EFI loader. If so, then we have also been
748 * passed the efi memmap, systab, etc., so we should use these data structures
749 * for initialization. Note, the efi init code path is determined by the
750 * global efi_enabled. This allows the same kernel image to be used on existing
751 * systems (with a traditional BIOS) as well as on EFI systems.
752 */
753 /*
754 * setup_arch - architecture-specific boot-time initializations
755 *
756 * Note: On x86_64, fixmaps are ready for use even before this is called.
757 */
758
setup_arch(char ** cmdline_p)759 void __init setup_arch(char **cmdline_p)
760 {
761 #ifdef CONFIG_X86_32
762 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
763 visws_early_detect();
764
765 /*
766 * copy kernel address range established so far and switch
767 * to the proper swapper page table
768 */
769 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
770 initial_page_table + KERNEL_PGD_BOUNDARY,
771 KERNEL_PGD_PTRS);
772
773 load_cr3(swapper_pg_dir);
774 __flush_tlb_all();
775 #else
776 printk(KERN_INFO "Command line: %s\n", boot_command_line);
777 #endif
778
779 /*
780 * If we have OLPC OFW, we might end up relocating the fixmap due to
781 * reserve_top(), so do this before touching the ioremap area.
782 */
783 olpc_ofw_detect();
784
785 early_trap_init();
786 early_cpu_init();
787 early_ioremap_init();
788
789 setup_olpc_ofw_pgd();
790
791 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
792 screen_info = boot_params.screen_info;
793 edid_info = boot_params.edid_info;
794 #ifdef CONFIG_X86_32
795 apm_info.bios = boot_params.apm_bios_info;
796 ist_info = boot_params.ist_info;
797 if (boot_params.sys_desc_table.length != 0) {
798 set_mca_bus(boot_params.sys_desc_table.table[3] & 0x2);
799 machine_id = boot_params.sys_desc_table.table[0];
800 machine_submodel_id = boot_params.sys_desc_table.table[1];
801 BIOS_revision = boot_params.sys_desc_table.table[2];
802 }
803 #endif
804 saved_video_mode = boot_params.hdr.vid_mode;
805 bootloader_type = boot_params.hdr.type_of_loader;
806 if ((bootloader_type >> 4) == 0xe) {
807 bootloader_type &= 0xf;
808 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
809 }
810 bootloader_version = bootloader_type & 0xf;
811 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
812
813 #ifdef CONFIG_BLK_DEV_RAM
814 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
815 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
816 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
817 #endif
818 #ifdef CONFIG_EFI
819 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
820 "EL32", 4)) {
821 set_bit(EFI_BOOT, &x86_efi_facility);
822 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
823 "EL64", 4)) {
824 set_bit(EFI_BOOT, &x86_efi_facility);
825 set_bit(EFI_64BIT, &x86_efi_facility);
826 }
827
828 if (efi_enabled(EFI_BOOT))
829 efi_memblock_x86_reserve_range();
830 #endif
831
832 x86_init.oem.arch_setup();
833
834 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
835 setup_memory_map();
836 parse_setup_data();
837 /* update the e820_saved too */
838 e820_reserve_setup_data();
839
840 copy_edd();
841
842 if (!boot_params.hdr.root_flags)
843 root_mountflags &= ~MS_RDONLY;
844 init_mm.start_code = (unsigned long) _text;
845 init_mm.end_code = (unsigned long) _etext;
846 init_mm.end_data = (unsigned long) _edata;
847 init_mm.brk = _brk_end;
848
849 code_resource.start = virt_to_phys(_text);
850 code_resource.end = virt_to_phys(_etext)-1;
851 data_resource.start = virt_to_phys(_etext);
852 data_resource.end = virt_to_phys(_edata)-1;
853 bss_resource.start = virt_to_phys(&__bss_start);
854 bss_resource.end = virt_to_phys(&__bss_stop)-1;
855
856 #ifdef CONFIG_CMDLINE_BOOL
857 #ifdef CONFIG_CMDLINE_OVERRIDE
858 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
859 #else
860 if (builtin_cmdline[0]) {
861 /* append boot loader cmdline to builtin */
862 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
863 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
864 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
865 }
866 #endif
867 #endif
868
869 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
870 *cmdline_p = command_line;
871
872 /*
873 * x86_configure_nx() is called before parse_early_param() to detect
874 * whether hardware doesn't support NX (so that the early EHCI debug
875 * console setup can safely call set_fixmap()). It may then be called
876 * again from within noexec_setup() during parsing early parameters
877 * to honor the respective command line option.
878 */
879 x86_configure_nx();
880
881 parse_early_param();
882
883 x86_report_nx();
884
885 /* after early param, so could get panic from serial */
886 memblock_x86_reserve_range_setup_data();
887
888 if (acpi_mps_check()) {
889 #ifdef CONFIG_X86_LOCAL_APIC
890 disable_apic = 1;
891 #endif
892 setup_clear_cpu_cap(X86_FEATURE_APIC);
893 }
894
895 #ifdef CONFIG_PCI
896 if (pci_early_dump_regs)
897 early_dump_pci_devices();
898 #endif
899
900 finish_e820_parsing();
901
902 if (efi_enabled(EFI_BOOT))
903 efi_init();
904
905 dmi_scan_machine();
906
907 /*
908 * VMware detection requires dmi to be available, so this
909 * needs to be done after dmi_scan_machine, for the BP.
910 */
911 init_hypervisor_platform();
912
913 x86_init.resources.probe_roms();
914
915 /* after parse_early_param, so could debug it */
916 insert_resource(&iomem_resource, &code_resource);
917 insert_resource(&iomem_resource, &data_resource);
918 insert_resource(&iomem_resource, &bss_resource);
919
920 trim_bios_range();
921 #ifdef CONFIG_X86_32
922 if (ppro_with_ram_bug()) {
923 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
924 E820_RESERVED);
925 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
926 printk(KERN_INFO "fixed physical RAM map:\n");
927 e820_print_map("bad_ppro");
928 }
929 #else
930 early_gart_iommu_check();
931 #endif
932
933 /*
934 * partially used pages are not usable - thus
935 * we are rounding upwards:
936 */
937 max_pfn = e820_end_of_ram_pfn();
938
939 /* update e820 for memory not covered by WB MTRRs */
940 mtrr_bp_init();
941 if (mtrr_trim_uncached_memory(max_pfn))
942 max_pfn = e820_end_of_ram_pfn();
943
944 #ifdef CONFIG_X86_32
945 /* max_low_pfn get updated here */
946 find_low_pfn_range();
947 #else
948 num_physpages = max_pfn;
949
950 check_x2apic();
951
952 /* How many end-of-memory variables you have, grandma! */
953 /* need this before calling reserve_initrd */
954 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
955 max_low_pfn = e820_end_of_low_ram_pfn();
956 else
957 max_low_pfn = max_pfn;
958
959 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
960 #endif
961
962 /*
963 * Find and reserve possible boot-time SMP configuration:
964 */
965 find_smp_config();
966
967 reserve_ibft_region();
968
969 /*
970 * Need to conclude brk, before memblock_x86_fill()
971 * it could use memblock_find_in_range, could overlap with
972 * brk area.
973 */
974 reserve_brk();
975
976 cleanup_highmap();
977
978 memblock.current_limit = get_max_mapped();
979 memblock_x86_fill();
980
981 /*
982 * The EFI specification says that boot service code won't be called
983 * after ExitBootServices(). This is, in fact, a lie.
984 */
985 if (efi_enabled(EFI_MEMMAP))
986 efi_reserve_boot_services();
987
988 /* preallocate 4k for mptable mpc */
989 early_reserve_e820_mpc_new();
990
991 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
992 setup_bios_corruption_check();
993 #endif
994
995 printk(KERN_DEBUG "initial memory mapped : 0 - %08lx\n",
996 max_pfn_mapped<<PAGE_SHIFT);
997
998 setup_trampolines();
999
1000 trim_platform_memory_ranges();
1001
1002 init_gbpages();
1003
1004 /* max_pfn_mapped is updated here */
1005 max_low_pfn_mapped = init_memory_mapping(0, max_low_pfn<<PAGE_SHIFT);
1006 max_pfn_mapped = max_low_pfn_mapped;
1007
1008 #ifdef CONFIG_X86_64
1009 if (max_pfn > max_low_pfn) {
1010 int i;
1011 unsigned long start, end;
1012 unsigned long start_pfn, end_pfn;
1013
1014 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn,
1015 NULL) {
1016
1017 end = PFN_PHYS(end_pfn);
1018 if (end <= (1UL<<32))
1019 continue;
1020
1021 start = PFN_PHYS(start_pfn);
1022 max_pfn_mapped = init_memory_mapping(
1023 max((1UL<<32), start), end);
1024 }
1025
1026 /* can we preseve max_low_pfn ?*/
1027 max_low_pfn = max_pfn;
1028 }
1029 #endif
1030 memblock.current_limit = get_max_mapped();
1031
1032 /*
1033 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1034 */
1035
1036 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1037 if (init_ohci1394_dma_early)
1038 init_ohci1394_dma_on_all_controllers();
1039 #endif
1040 /* Allocate bigger log buffer */
1041 setup_log_buf(1);
1042
1043 reserve_initrd();
1044
1045 reserve_crashkernel();
1046
1047 vsmp_init();
1048
1049 io_delay_init();
1050
1051 /*
1052 * Parse the ACPI tables for possible boot-time SMP configuration.
1053 */
1054 acpi_boot_table_init();
1055
1056 early_acpi_boot_init();
1057
1058 initmem_init();
1059 memblock_find_dma_reserve();
1060
1061 #ifdef CONFIG_KVM_CLOCK
1062 kvmclock_init();
1063 #endif
1064
1065 x86_init.paging.pagetable_setup_start(swapper_pg_dir);
1066 paging_init();
1067 x86_init.paging.pagetable_setup_done(swapper_pg_dir);
1068
1069 if (boot_cpu_data.cpuid_level >= 0) {
1070 /* A CPU has %cr4 if and only if it has CPUID */
1071 mmu_cr4_features = read_cr4();
1072 }
1073
1074 #ifdef CONFIG_X86_32
1075 /* sync back kernel address range */
1076 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1077 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1078 KERNEL_PGD_PTRS);
1079 #endif
1080
1081 tboot_probe();
1082
1083 #ifdef CONFIG_X86_64
1084 map_vsyscall();
1085 #endif
1086
1087 generic_apic_probe();
1088
1089 early_quirks();
1090
1091 /*
1092 * Read APIC and some other early information from ACPI tables.
1093 */
1094 acpi_boot_init();
1095 sfi_init();
1096 x86_dtb_init();
1097
1098 /*
1099 * get boot-time SMP configuration:
1100 */
1101 if (smp_found_config)
1102 get_smp_config();
1103
1104 prefill_possible_map();
1105
1106 init_cpu_to_node();
1107
1108 init_apic_mappings();
1109 ioapic_and_gsi_init();
1110
1111 kvm_guest_init();
1112
1113 e820_reserve_resources();
1114 e820_mark_nosave_regions(max_low_pfn);
1115
1116 x86_init.resources.reserve_resources();
1117
1118 e820_setup_gap();
1119
1120 #ifdef CONFIG_VT
1121 #if defined(CONFIG_VGA_CONSOLE)
1122 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1123 conswitchp = &vga_con;
1124 #elif defined(CONFIG_DUMMY_CONSOLE)
1125 conswitchp = &dummy_con;
1126 #endif
1127 #endif
1128 x86_init.oem.banner();
1129
1130 x86_init.timers.wallclock_init();
1131
1132 x86_platform.wallclock_init();
1133
1134 mcheck_init();
1135
1136 arch_init_ideal_nops();
1137
1138 #ifdef CONFIG_EFI
1139 /* Once setup is done above, unmap the EFI memory map on
1140 * mismatched firmware/kernel archtectures since there is no
1141 * support for runtime services.
1142 */
1143 if (efi_enabled(EFI_BOOT) &&
1144 IS_ENABLED(CONFIG_X86_64) != efi_enabled(EFI_64BIT)) {
1145 pr_info("efi: Setup done, disabling due to 32/64-bit mismatch\n");
1146 efi_unmap_memmap();
1147 }
1148 #endif
1149 }
1150
1151 #ifdef CONFIG_X86_32
1152
1153 static struct resource video_ram_resource = {
1154 .name = "Video RAM area",
1155 .start = 0xa0000,
1156 .end = 0xbffff,
1157 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1158 };
1159
i386_reserve_resources(void)1160 void __init i386_reserve_resources(void)
1161 {
1162 request_resource(&iomem_resource, &video_ram_resource);
1163 reserve_standard_io_resources();
1164 }
1165
1166 #endif /* CONFIG_X86_32 */
1167