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/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/iscsi_ibft.h>
44 #include <linux/nodemask.h>
45 #include <linux/kexec.h>
46 #include <linux/dmi.h>
47 #include <linux/pfn.h>
48 #include <linux/pci.h>
49 #include <asm/pci-direct.h>
50 #include <linux/init_ohci1394_dma.h>
51 #include <linux/kvm_para.h>
52 #include <linux/dma-contiguous.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 #include <linux/jiffies.h>
72
73 #include <video/edid.h>
74
75 #include <asm/mtrr.h>
76 #include <asm/apic.h>
77 #include <asm/realmode.h>
78 #include <asm/e820.h>
79 #include <asm/mpspec.h>
80 #include <asm/setup.h>
81 #include <asm/efi.h>
82 #include <asm/timer.h>
83 #include <asm/i8259.h>
84 #include <asm/sections.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 #include <asm/mce.h>
111 #include <asm/alternative.h>
112 #include <asm/prom.h>
113
114 /*
115 * max_low_pfn_mapped: highest direct mapped pfn under 4GB
116 * max_pfn_mapped: highest direct mapped pfn over 4GB
117 *
118 * The direct mapping only covers E820_RAM regions, so the ranges and gaps are
119 * represented by pfn_mapped
120 */
121 unsigned long max_low_pfn_mapped;
122 unsigned long max_pfn_mapped;
123
124 #ifdef CONFIG_DMI
125 RESERVE_BRK(dmi_alloc, 65536);
126 #endif
127
128
129 static __initdata unsigned long _brk_start = (unsigned long)__brk_base;
130 unsigned long _brk_end = (unsigned long)__brk_base;
131
132 #ifdef CONFIG_X86_64
default_cpu_present_to_apicid(int mps_cpu)133 int default_cpu_present_to_apicid(int mps_cpu)
134 {
135 return __default_cpu_present_to_apicid(mps_cpu);
136 }
137
default_check_phys_apicid_present(int phys_apicid)138 int default_check_phys_apicid_present(int phys_apicid)
139 {
140 return __default_check_phys_apicid_present(phys_apicid);
141 }
142 #endif
143
144 struct boot_params boot_params;
145
146 /*
147 * Machine setup..
148 */
149 static struct resource data_resource = {
150 .name = "Kernel data",
151 .start = 0,
152 .end = 0,
153 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
154 };
155
156 static struct resource code_resource = {
157 .name = "Kernel code",
158 .start = 0,
159 .end = 0,
160 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
161 };
162
163 static struct resource bss_resource = {
164 .name = "Kernel bss",
165 .start = 0,
166 .end = 0,
167 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
168 };
169
170
171 #ifdef CONFIG_X86_32
172 /* cpu data as detected by the assembly code in head.S */
173 struct cpuinfo_x86 new_cpu_data __cpuinitdata = {
174 .wp_works_ok = -1,
175 };
176 /* common cpu data for all cpus */
177 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
178 .wp_works_ok = -1,
179 };
180 EXPORT_SYMBOL(boot_cpu_data);
181
182 unsigned int def_to_bigsmp;
183
184 /* for MCA, but anyone else can use it if they want */
185 unsigned int machine_id;
186 unsigned int machine_submodel_id;
187 unsigned int BIOS_revision;
188
189 struct apm_info apm_info;
190 EXPORT_SYMBOL(apm_info);
191
192 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
193 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
194 struct ist_info ist_info;
195 EXPORT_SYMBOL(ist_info);
196 #else
197 struct ist_info ist_info;
198 #endif
199
200 #else
201 struct cpuinfo_x86 boot_cpu_data __read_mostly = {
202 .x86_phys_bits = MAX_PHYSMEM_BITS,
203 };
204 EXPORT_SYMBOL(boot_cpu_data);
205 #endif
206
207
208 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
209 unsigned long mmu_cr4_features;
210 #else
211 unsigned long mmu_cr4_features = X86_CR4_PAE;
212 #endif
213
214 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
215 int bootloader_type, bootloader_version;
216
217 /*
218 * Setup options
219 */
220 struct screen_info screen_info;
221 EXPORT_SYMBOL(screen_info);
222 struct edid_info edid_info;
223 EXPORT_SYMBOL_GPL(edid_info);
224
225 extern int root_mountflags;
226
227 unsigned long saved_video_mode;
228
229 #define RAMDISK_IMAGE_START_MASK 0x07FF
230 #define RAMDISK_PROMPT_FLAG 0x8000
231 #define RAMDISK_LOAD_FLAG 0x4000
232
233 static char __initdata command_line[COMMAND_LINE_SIZE];
234 #ifdef CONFIG_CMDLINE_BOOL
235 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
236 #endif
237
238 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
239 struct edd edd;
240 #ifdef CONFIG_EDD_MODULE
241 EXPORT_SYMBOL(edd);
242 #endif
243 /**
244 * copy_edd() - Copy the BIOS EDD information
245 * from boot_params into a safe place.
246 *
247 */
copy_edd(void)248 static inline void __init copy_edd(void)
249 {
250 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
251 sizeof(edd.mbr_signature));
252 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
253 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
254 edd.edd_info_nr = boot_params.eddbuf_entries;
255 }
256 #else
copy_edd(void)257 static inline void __init copy_edd(void)
258 {
259 }
260 #endif
261
extend_brk(size_t size,size_t align)262 void * __init extend_brk(size_t size, size_t align)
263 {
264 size_t mask = align - 1;
265 void *ret;
266
267 BUG_ON(_brk_start == 0);
268 BUG_ON(align & mask);
269
270 _brk_end = (_brk_end + mask) & ~mask;
271 BUG_ON((char *)(_brk_end + size) > __brk_limit);
272
273 ret = (void *)_brk_end;
274 _brk_end += size;
275
276 memset(ret, 0, size);
277
278 return ret;
279 }
280
281 #ifdef CONFIG_X86_32
cleanup_highmap(void)282 static void __init cleanup_highmap(void)
283 {
284 }
285 #endif
286
reserve_brk(void)287 static void __init reserve_brk(void)
288 {
289 if (_brk_end > _brk_start)
290 memblock_reserve(__pa_symbol(_brk_start),
291 _brk_end - _brk_start);
292
293 /* Mark brk area as locked down and no longer taking any
294 new allocations */
295 _brk_start = 0;
296 }
297
298 #ifdef CONFIG_BLK_DEV_INITRD
299
get_ramdisk_image(void)300 static u64 __init get_ramdisk_image(void)
301 {
302 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
303
304 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
305
306 return ramdisk_image;
307 }
get_ramdisk_size(void)308 static u64 __init get_ramdisk_size(void)
309 {
310 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
311
312 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
313
314 return ramdisk_size;
315 }
316
317 #define MAX_MAP_CHUNK (NR_FIX_BTMAPS << PAGE_SHIFT)
relocate_initrd(void)318 static void __init relocate_initrd(void)
319 {
320 /* Assume only end is not page aligned */
321 u64 ramdisk_image = get_ramdisk_image();
322 u64 ramdisk_size = get_ramdisk_size();
323 u64 area_size = PAGE_ALIGN(ramdisk_size);
324 u64 ramdisk_here;
325 unsigned long slop, clen, mapaddr;
326 char *p, *q;
327
328 /* We need to move the initrd down into directly mapped mem */
329 ramdisk_here = memblock_find_in_range(0, PFN_PHYS(max_pfn_mapped),
330 area_size, PAGE_SIZE);
331
332 if (!ramdisk_here)
333 panic("Cannot find place for new RAMDISK of size %lld\n",
334 ramdisk_size);
335
336 /* Note: this includes all the mem currently occupied by
337 the initrd, we rely on that fact to keep the data intact. */
338 memblock_reserve(ramdisk_here, area_size);
339 initrd_start = ramdisk_here + PAGE_OFFSET;
340 initrd_end = initrd_start + ramdisk_size;
341 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
342 ramdisk_here, ramdisk_here + ramdisk_size - 1);
343
344 q = (char *)initrd_start;
345
346 /* Copy the initrd */
347 while (ramdisk_size) {
348 slop = ramdisk_image & ~PAGE_MASK;
349 clen = ramdisk_size;
350 if (clen > MAX_MAP_CHUNK-slop)
351 clen = MAX_MAP_CHUNK-slop;
352 mapaddr = ramdisk_image & PAGE_MASK;
353 p = early_memremap(mapaddr, clen+slop);
354 memcpy(q, p+slop, clen);
355 early_iounmap(p, clen+slop);
356 q += clen;
357 ramdisk_image += clen;
358 ramdisk_size -= clen;
359 }
360
361 ramdisk_image = get_ramdisk_image();
362 ramdisk_size = get_ramdisk_size();
363 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
364 " [mem %#010llx-%#010llx]\n",
365 ramdisk_image, ramdisk_image + ramdisk_size - 1,
366 ramdisk_here, ramdisk_here + ramdisk_size - 1);
367 }
368
early_reserve_initrd(void)369 static void __init early_reserve_initrd(void)
370 {
371 /* Assume only end is not page aligned */
372 u64 ramdisk_image = get_ramdisk_image();
373 u64 ramdisk_size = get_ramdisk_size();
374 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
375
376 if (!boot_params.hdr.type_of_loader ||
377 !ramdisk_image || !ramdisk_size)
378 return; /* No initrd provided by bootloader */
379
380 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
381 }
reserve_initrd(void)382 static void __init reserve_initrd(void)
383 {
384 /* Assume only end is not page aligned */
385 u64 ramdisk_image = get_ramdisk_image();
386 u64 ramdisk_size = get_ramdisk_size();
387 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
388 u64 mapped_size;
389
390 if (!boot_params.hdr.type_of_loader ||
391 !ramdisk_image || !ramdisk_size)
392 return; /* No initrd provided by bootloader */
393
394 initrd_start = 0;
395
396 mapped_size = memblock_mem_size(max_pfn_mapped);
397 if (ramdisk_size >= (mapped_size>>1))
398 panic("initrd too large to handle, "
399 "disabling initrd (%lld needed, %lld available)\n",
400 ramdisk_size, mapped_size>>1);
401
402 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
403 ramdisk_end - 1);
404
405 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
406 PFN_DOWN(ramdisk_end))) {
407 /* All are mapped, easy case */
408 initrd_start = ramdisk_image + PAGE_OFFSET;
409 initrd_end = initrd_start + ramdisk_size;
410 return;
411 }
412
413 relocate_initrd();
414
415 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
416 }
417 #else
early_reserve_initrd(void)418 static void __init early_reserve_initrd(void)
419 {
420 }
reserve_initrd(void)421 static void __init reserve_initrd(void)
422 {
423 }
424 #endif /* CONFIG_BLK_DEV_INITRD */
425
parse_setup_data(void)426 static void __init parse_setup_data(void)
427 {
428 struct setup_data *data;
429 u64 pa_data;
430
431 pa_data = boot_params.hdr.setup_data;
432 while (pa_data) {
433 u32 data_len, map_len;
434
435 map_len = max(PAGE_SIZE - (pa_data & ~PAGE_MASK),
436 (u64)sizeof(struct setup_data));
437 data = early_memremap(pa_data, map_len);
438 data_len = data->len + sizeof(struct setup_data);
439 if (data_len > map_len) {
440 early_iounmap(data, map_len);
441 data = early_memremap(pa_data, data_len);
442 map_len = data_len;
443 }
444
445 switch (data->type) {
446 case SETUP_E820_EXT:
447 parse_e820_ext(data);
448 break;
449 case SETUP_DTB:
450 add_dtb(pa_data);
451 break;
452 default:
453 break;
454 }
455 pa_data = data->next;
456 early_iounmap(data, map_len);
457 }
458 }
459
e820_reserve_setup_data(void)460 static void __init e820_reserve_setup_data(void)
461 {
462 struct setup_data *data;
463 u64 pa_data;
464 int found = 0;
465
466 pa_data = boot_params.hdr.setup_data;
467 while (pa_data) {
468 data = early_memremap(pa_data, sizeof(*data));
469 e820_update_range(pa_data, sizeof(*data)+data->len,
470 E820_RAM, E820_RESERVED_KERN);
471 found = 1;
472 pa_data = data->next;
473 early_iounmap(data, sizeof(*data));
474 }
475 if (!found)
476 return;
477
478 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
479 memcpy(&e820_saved, &e820, sizeof(struct e820map));
480 printk(KERN_INFO "extended physical RAM map:\n");
481 e820_print_map("reserve setup_data");
482 }
483
memblock_x86_reserve_range_setup_data(void)484 static void __init memblock_x86_reserve_range_setup_data(void)
485 {
486 struct setup_data *data;
487 u64 pa_data;
488
489 pa_data = boot_params.hdr.setup_data;
490 while (pa_data) {
491 data = early_memremap(pa_data, sizeof(*data));
492 memblock_reserve(pa_data, sizeof(*data) + data->len);
493 pa_data = data->next;
494 early_iounmap(data, sizeof(*data));
495 }
496 }
497
498 /*
499 * --------- Crashkernel reservation ------------------------------
500 */
501
502 #ifdef CONFIG_KEXEC
503
504 /*
505 * Keep the crash kernel below this limit. On 32 bits earlier kernels
506 * would limit the kernel to the low 512 MiB due to mapping restrictions.
507 * On 64bit, old kexec-tools need to under 896MiB.
508 */
509 #ifdef CONFIG_X86_32
510 # define CRASH_KERNEL_ADDR_LOW_MAX (512 << 20)
511 # define CRASH_KERNEL_ADDR_HIGH_MAX (512 << 20)
512 #else
513 # define CRASH_KERNEL_ADDR_LOW_MAX (896UL<<20)
514 # define CRASH_KERNEL_ADDR_HIGH_MAX MAXMEM
515 #endif
516
reserve_crashkernel_low(void)517 static void __init reserve_crashkernel_low(void)
518 {
519 #ifdef CONFIG_X86_64
520 const unsigned long long alignment = 16<<20; /* 16M */
521 unsigned long long low_base = 0, low_size = 0;
522 unsigned long total_low_mem;
523 unsigned long long base;
524 bool auto_set = false;
525 int ret;
526
527 total_low_mem = memblock_mem_size(1UL<<(32-PAGE_SHIFT));
528 /* crashkernel=Y,low */
529 ret = parse_crashkernel_low(boot_command_line, total_low_mem,
530 &low_size, &base);
531 if (ret != 0) {
532 /*
533 * two parts from lib/swiotlb.c:
534 * swiotlb size: user specified with swiotlb= or default.
535 * swiotlb overflow buffer: now is hardcoded to 32k.
536 * We round it to 8M for other buffers that
537 * may need to stay low too.
538 */
539 low_size = swiotlb_size_or_default() + (8UL<<20);
540 auto_set = true;
541 } else {
542 /* passed with crashkernel=0,low ? */
543 if (!low_size)
544 return;
545 }
546
547 low_base = memblock_find_in_range(low_size, (1ULL<<32),
548 low_size, alignment);
549
550 if (!low_base) {
551 if (!auto_set)
552 pr_info("crashkernel low reservation failed - No suitable area found.\n");
553
554 return;
555 }
556
557 memblock_reserve(low_base, low_size);
558 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (System low RAM: %ldMB)\n",
559 (unsigned long)(low_size >> 20),
560 (unsigned long)(low_base >> 20),
561 (unsigned long)(total_low_mem >> 20));
562 crashk_low_res.start = low_base;
563 crashk_low_res.end = low_base + low_size - 1;
564 insert_resource(&iomem_resource, &crashk_low_res);
565 #endif
566 }
567
reserve_crashkernel(void)568 static void __init reserve_crashkernel(void)
569 {
570 const unsigned long long alignment = 16<<20; /* 16M */
571 unsigned long long total_mem;
572 unsigned long long crash_size, crash_base;
573 bool high = false;
574 int ret;
575
576 total_mem = memblock_phys_mem_size();
577
578 /* crashkernel=XM */
579 ret = parse_crashkernel(boot_command_line, total_mem,
580 &crash_size, &crash_base);
581 if (ret != 0 || crash_size <= 0) {
582 /* crashkernel=X,high */
583 ret = parse_crashkernel_high(boot_command_line, total_mem,
584 &crash_size, &crash_base);
585 if (ret != 0 || crash_size <= 0)
586 return;
587 high = true;
588 }
589
590 /* 0 means: find the address automatically */
591 if (crash_base <= 0) {
592 /*
593 * kexec want bzImage is below CRASH_KERNEL_ADDR_MAX
594 */
595 crash_base = memblock_find_in_range(alignment,
596 high ? CRASH_KERNEL_ADDR_HIGH_MAX :
597 CRASH_KERNEL_ADDR_LOW_MAX,
598 crash_size, alignment);
599
600 if (!crash_base) {
601 pr_info("crashkernel reservation failed - No suitable area found.\n");
602 return;
603 }
604
605 } else {
606 unsigned long long start;
607
608 start = memblock_find_in_range(crash_base,
609 crash_base + crash_size, crash_size, 1<<20);
610 if (start != crash_base) {
611 pr_info("crashkernel reservation failed - memory is in use.\n");
612 return;
613 }
614 }
615 memblock_reserve(crash_base, crash_size);
616
617 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
618 "for crashkernel (System RAM: %ldMB)\n",
619 (unsigned long)(crash_size >> 20),
620 (unsigned long)(crash_base >> 20),
621 (unsigned long)(total_mem >> 20));
622
623 crashk_res.start = crash_base;
624 crashk_res.end = crash_base + crash_size - 1;
625 insert_resource(&iomem_resource, &crashk_res);
626
627 if (crash_base >= (1ULL<<32))
628 reserve_crashkernel_low();
629 }
630 #else
reserve_crashkernel(void)631 static void __init reserve_crashkernel(void)
632 {
633 }
634 #endif
635
636 static struct resource standard_io_resources[] = {
637 { .name = "dma1", .start = 0x00, .end = 0x1f,
638 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
639 { .name = "pic1", .start = 0x20, .end = 0x21,
640 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
641 { .name = "timer0", .start = 0x40, .end = 0x43,
642 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
643 { .name = "timer1", .start = 0x50, .end = 0x53,
644 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
645 { .name = "keyboard", .start = 0x60, .end = 0x60,
646 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
647 { .name = "keyboard", .start = 0x64, .end = 0x64,
648 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
649 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
650 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
651 { .name = "pic2", .start = 0xa0, .end = 0xa1,
652 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
653 { .name = "dma2", .start = 0xc0, .end = 0xdf,
654 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
655 { .name = "fpu", .start = 0xf0, .end = 0xff,
656 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
657 };
658
reserve_standard_io_resources(void)659 void __init reserve_standard_io_resources(void)
660 {
661 int i;
662
663 /* request I/O space for devices used on all i[345]86 PCs */
664 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
665 request_resource(&ioport_resource, &standard_io_resources[i]);
666
667 }
668
reserve_ibft_region(void)669 static __init void reserve_ibft_region(void)
670 {
671 unsigned long addr, size = 0;
672
673 addr = find_ibft_region(&size);
674
675 if (size)
676 memblock_reserve(addr, size);
677 }
678
snb_gfx_workaround_needed(void)679 static bool __init snb_gfx_workaround_needed(void)
680 {
681 #ifdef CONFIG_PCI
682 int i;
683 u16 vendor, devid;
684 static const __initconst u16 snb_ids[] = {
685 0x0102,
686 0x0112,
687 0x0122,
688 0x0106,
689 0x0116,
690 0x0126,
691 0x010a,
692 };
693
694 /* Assume no if something weird is going on with PCI */
695 if (!early_pci_allowed())
696 return false;
697
698 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
699 if (vendor != 0x8086)
700 return false;
701
702 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
703 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
704 if (devid == snb_ids[i])
705 return true;
706 #endif
707
708 return false;
709 }
710
711 /*
712 * Sandy Bridge graphics has trouble with certain ranges, exclude
713 * them from allocation.
714 */
trim_snb_memory(void)715 static void __init trim_snb_memory(void)
716 {
717 static const __initconst unsigned long bad_pages[] = {
718 0x20050000,
719 0x20110000,
720 0x20130000,
721 0x20138000,
722 0x40004000,
723 };
724 int i;
725
726 if (!snb_gfx_workaround_needed())
727 return;
728
729 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
730
731 /*
732 * Reserve all memory below the 1 MB mark that has not
733 * already been reserved.
734 */
735 memblock_reserve(0, 1<<20);
736
737 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
738 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
739 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
740 bad_pages[i]);
741 }
742 }
743
744 /*
745 * Here we put platform-specific memory range workarounds, i.e.
746 * memory known to be corrupt or otherwise in need to be reserved on
747 * specific platforms.
748 *
749 * If this gets used more widely it could use a real dispatch mechanism.
750 */
trim_platform_memory_ranges(void)751 static void __init trim_platform_memory_ranges(void)
752 {
753 trim_snb_memory();
754 }
755
trim_bios_range(void)756 static void __init trim_bios_range(void)
757 {
758 /*
759 * A special case is the first 4Kb of memory;
760 * This is a BIOS owned area, not kernel ram, but generally
761 * not listed as such in the E820 table.
762 *
763 * This typically reserves additional memory (64KiB by default)
764 * since some BIOSes are known to corrupt low memory. See the
765 * Kconfig help text for X86_RESERVE_LOW.
766 */
767 e820_update_range(0, PAGE_SIZE, E820_RAM, E820_RESERVED);
768
769 /*
770 * special case: Some BIOSen report the PC BIOS
771 * area (640->1Mb) as ram even though it is not.
772 * take them out.
773 */
774 e820_remove_range(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_RAM, 1);
775
776 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
777 }
778
779 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)780 static void __init e820_add_kernel_range(void)
781 {
782 u64 start = __pa_symbol(_text);
783 u64 size = __pa_symbol(_end) - start;
784
785 /*
786 * Complain if .text .data and .bss are not marked as E820_RAM and
787 * attempt to fix it by adding the range. We may have a confused BIOS,
788 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
789 * exclude kernel range. If we really are running on top non-RAM,
790 * we will crash later anyways.
791 */
792 if (e820_all_mapped(start, start + size, E820_RAM))
793 return;
794
795 pr_warn(".text .data .bss are not marked as E820_RAM!\n");
796 e820_remove_range(start, size, E820_RAM, 0);
797 e820_add_region(start, size, E820_RAM);
798 }
799
800 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
801
parse_reservelow(char * p)802 static int __init parse_reservelow(char *p)
803 {
804 unsigned long long size;
805
806 if (!p)
807 return -EINVAL;
808
809 size = memparse(p, &p);
810
811 if (size < 4096)
812 size = 4096;
813
814 if (size > 640*1024)
815 size = 640*1024;
816
817 reserve_low = size;
818
819 return 0;
820 }
821
822 early_param("reservelow", parse_reservelow);
823
trim_low_memory_range(void)824 static void __init trim_low_memory_range(void)
825 {
826 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
827 }
828
829 /*
830 * Determine if we were loaded by an EFI loader. If so, then we have also been
831 * passed the efi memmap, systab, etc., so we should use these data structures
832 * for initialization. Note, the efi init code path is determined by the
833 * global efi_enabled. This allows the same kernel image to be used on existing
834 * systems (with a traditional BIOS) as well as on EFI systems.
835 */
836 /*
837 * setup_arch - architecture-specific boot-time initializations
838 *
839 * Note: On x86_64, fixmaps are ready for use even before this is called.
840 */
841
setup_arch(char ** cmdline_p)842 void __init setup_arch(char **cmdline_p)
843 {
844 memblock_reserve(__pa_symbol(_text),
845 (unsigned long)__bss_stop - (unsigned long)_text);
846
847 early_reserve_initrd();
848
849 /*
850 * At this point everything still needed from the boot loader
851 * or BIOS or kernel text should be early reserved or marked not
852 * RAM in e820. All other memory is free game.
853 */
854
855 #ifdef CONFIG_X86_32
856 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
857 visws_early_detect();
858
859 /*
860 * copy kernel address range established so far and switch
861 * to the proper swapper page table
862 */
863 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
864 initial_page_table + KERNEL_PGD_BOUNDARY,
865 KERNEL_PGD_PTRS);
866
867 load_cr3(swapper_pg_dir);
868 __flush_tlb_all();
869 #else
870 printk(KERN_INFO "Command line: %s\n", boot_command_line);
871 #endif
872
873 /*
874 * If we have OLPC OFW, we might end up relocating the fixmap due to
875 * reserve_top(), so do this before touching the ioremap area.
876 */
877 olpc_ofw_detect();
878
879 early_trap_init();
880 early_cpu_init();
881 early_ioremap_init();
882
883 setup_olpc_ofw_pgd();
884
885 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
886 screen_info = boot_params.screen_info;
887 edid_info = boot_params.edid_info;
888 #ifdef CONFIG_X86_32
889 apm_info.bios = boot_params.apm_bios_info;
890 ist_info = boot_params.ist_info;
891 if (boot_params.sys_desc_table.length != 0) {
892 machine_id = boot_params.sys_desc_table.table[0];
893 machine_submodel_id = boot_params.sys_desc_table.table[1];
894 BIOS_revision = boot_params.sys_desc_table.table[2];
895 }
896 #endif
897 saved_video_mode = boot_params.hdr.vid_mode;
898 bootloader_type = boot_params.hdr.type_of_loader;
899 if ((bootloader_type >> 4) == 0xe) {
900 bootloader_type &= 0xf;
901 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
902 }
903 bootloader_version = bootloader_type & 0xf;
904 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
905
906 #ifdef CONFIG_BLK_DEV_RAM
907 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
908 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
909 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
910 #endif
911 #ifdef CONFIG_EFI
912 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
913 "EL32", 4)) {
914 set_bit(EFI_BOOT, &x86_efi_facility);
915 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
916 "EL64", 4)) {
917 set_bit(EFI_BOOT, &x86_efi_facility);
918 set_bit(EFI_64BIT, &x86_efi_facility);
919 }
920
921 if (efi_enabled(EFI_BOOT))
922 efi_memblock_x86_reserve_range();
923 #endif
924
925 x86_init.oem.arch_setup();
926
927 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
928 setup_memory_map();
929 parse_setup_data();
930 /* update the e820_saved too */
931 e820_reserve_setup_data();
932
933 copy_edd();
934
935 if (!boot_params.hdr.root_flags)
936 root_mountflags &= ~MS_RDONLY;
937 init_mm.start_code = (unsigned long) _text;
938 init_mm.end_code = (unsigned long) _etext;
939 init_mm.end_data = (unsigned long) _edata;
940 init_mm.brk = _brk_end;
941
942 code_resource.start = __pa_symbol(_text);
943 code_resource.end = __pa_symbol(_etext)-1;
944 data_resource.start = __pa_symbol(_etext);
945 data_resource.end = __pa_symbol(_edata)-1;
946 bss_resource.start = __pa_symbol(__bss_start);
947 bss_resource.end = __pa_symbol(__bss_stop)-1;
948
949 #ifdef CONFIG_CMDLINE_BOOL
950 #ifdef CONFIG_CMDLINE_OVERRIDE
951 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
952 #else
953 if (builtin_cmdline[0]) {
954 /* append boot loader cmdline to builtin */
955 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
956 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
957 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
958 }
959 #endif
960 #endif
961
962 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
963 *cmdline_p = command_line;
964
965 /*
966 * x86_configure_nx() is called before parse_early_param() to detect
967 * whether hardware doesn't support NX (so that the early EHCI debug
968 * console setup can safely call set_fixmap()). It may then be called
969 * again from within noexec_setup() during parsing early parameters
970 * to honor the respective command line option.
971 */
972 x86_configure_nx();
973
974 parse_early_param();
975
976 x86_report_nx();
977
978 /* after early param, so could get panic from serial */
979 memblock_x86_reserve_range_setup_data();
980
981 if (acpi_mps_check()) {
982 #ifdef CONFIG_X86_LOCAL_APIC
983 disable_apic = 1;
984 #endif
985 setup_clear_cpu_cap(X86_FEATURE_APIC);
986 }
987
988 #ifdef CONFIG_PCI
989 if (pci_early_dump_regs)
990 early_dump_pci_devices();
991 #endif
992
993 finish_e820_parsing();
994
995 if (efi_enabled(EFI_BOOT))
996 efi_init();
997
998 dmi_scan_machine();
999 dmi_set_dump_stack_arch_desc();
1000
1001 /*
1002 * VMware detection requires dmi to be available, so this
1003 * needs to be done after dmi_scan_machine, for the BP.
1004 */
1005 init_hypervisor_platform();
1006
1007 x86_init.resources.probe_roms();
1008
1009 /* after parse_early_param, so could debug it */
1010 insert_resource(&iomem_resource, &code_resource);
1011 insert_resource(&iomem_resource, &data_resource);
1012 insert_resource(&iomem_resource, &bss_resource);
1013
1014 e820_add_kernel_range();
1015 trim_bios_range();
1016 #ifdef CONFIG_X86_32
1017 if (ppro_with_ram_bug()) {
1018 e820_update_range(0x70000000ULL, 0x40000ULL, E820_RAM,
1019 E820_RESERVED);
1020 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
1021 printk(KERN_INFO "fixed physical RAM map:\n");
1022 e820_print_map("bad_ppro");
1023 }
1024 #else
1025 early_gart_iommu_check();
1026 #endif
1027
1028 /*
1029 * partially used pages are not usable - thus
1030 * we are rounding upwards:
1031 */
1032 max_pfn = e820_end_of_ram_pfn();
1033
1034 /* update e820 for memory not covered by WB MTRRs */
1035 mtrr_bp_init();
1036 if (mtrr_trim_uncached_memory(max_pfn))
1037 max_pfn = e820_end_of_ram_pfn();
1038
1039 #ifdef CONFIG_X86_32
1040 /* max_low_pfn get updated here */
1041 find_low_pfn_range();
1042 #else
1043 num_physpages = max_pfn;
1044
1045 check_x2apic();
1046
1047 /* How many end-of-memory variables you have, grandma! */
1048 /* need this before calling reserve_initrd */
1049 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1050 max_low_pfn = e820_end_of_low_ram_pfn();
1051 else
1052 max_low_pfn = max_pfn;
1053
1054 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1055 #endif
1056
1057 /*
1058 * Find and reserve possible boot-time SMP configuration:
1059 */
1060 find_smp_config();
1061
1062 reserve_ibft_region();
1063
1064 early_alloc_pgt_buf();
1065
1066 /*
1067 * Need to conclude brk, before memblock_x86_fill()
1068 * it could use memblock_find_in_range, could overlap with
1069 * brk area.
1070 */
1071 reserve_brk();
1072
1073 cleanup_highmap();
1074
1075 memblock.current_limit = ISA_END_ADDRESS;
1076 memblock_x86_fill();
1077
1078 /*
1079 * The EFI specification says that boot service code won't be called
1080 * after ExitBootServices(). This is, in fact, a lie.
1081 */
1082 if (efi_enabled(EFI_MEMMAP))
1083 efi_reserve_boot_services();
1084
1085 /* preallocate 4k for mptable mpc */
1086 early_reserve_e820_mpc_new();
1087
1088 #ifdef CONFIG_X86_CHECK_BIOS_CORRUPTION
1089 setup_bios_corruption_check();
1090 #endif
1091
1092 #ifdef CONFIG_X86_32
1093 printk(KERN_DEBUG "initial memory mapped: [mem 0x00000000-%#010lx]\n",
1094 (max_pfn_mapped<<PAGE_SHIFT) - 1);
1095 #endif
1096
1097 reserve_real_mode();
1098
1099 trim_platform_memory_ranges();
1100 trim_low_memory_range();
1101
1102 init_mem_mapping();
1103
1104 early_trap_pf_init();
1105
1106 setup_real_mode();
1107
1108 memblock.current_limit = get_max_mapped();
1109 dma_contiguous_reserve(0);
1110
1111 /*
1112 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1113 */
1114
1115 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1116 if (init_ohci1394_dma_early)
1117 init_ohci1394_dma_on_all_controllers();
1118 #endif
1119 /* Allocate bigger log buffer */
1120 setup_log_buf(1);
1121
1122 reserve_initrd();
1123
1124 #if defined(CONFIG_ACPI) && defined(CONFIG_BLK_DEV_INITRD)
1125 acpi_initrd_override((void *)initrd_start, initrd_end - initrd_start);
1126 #endif
1127
1128 reserve_crashkernel();
1129
1130 vsmp_init();
1131
1132 io_delay_init();
1133
1134 /*
1135 * Parse the ACPI tables for possible boot-time SMP configuration.
1136 */
1137 acpi_boot_table_init();
1138
1139 early_acpi_boot_init();
1140
1141 initmem_init();
1142 memblock_find_dma_reserve();
1143
1144 #ifdef CONFIG_KVM_GUEST
1145 kvmclock_init();
1146 #endif
1147
1148 x86_init.paging.pagetable_init();
1149
1150 if (boot_cpu_data.cpuid_level >= 0) {
1151 /* A CPU has %cr4 if and only if it has CPUID */
1152 mmu_cr4_features = read_cr4();
1153 if (trampoline_cr4_features)
1154 *trampoline_cr4_features = mmu_cr4_features;
1155 }
1156
1157 #ifdef CONFIG_X86_32
1158 /* sync back kernel address range */
1159 clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
1160 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
1161 KERNEL_PGD_PTRS);
1162 #endif
1163
1164 tboot_probe();
1165
1166 #ifdef CONFIG_X86_64
1167 map_vsyscall();
1168 #endif
1169
1170 generic_apic_probe();
1171
1172 early_quirks();
1173
1174 /*
1175 * Read APIC and some other early information from ACPI tables.
1176 */
1177 acpi_boot_init();
1178 sfi_init();
1179 x86_dtb_init();
1180
1181 /*
1182 * get boot-time SMP configuration:
1183 */
1184 if (smp_found_config)
1185 get_smp_config();
1186
1187 prefill_possible_map();
1188
1189 init_cpu_to_node();
1190
1191 init_apic_mappings();
1192 if (x86_io_apic_ops.init)
1193 x86_io_apic_ops.init();
1194
1195 kvm_guest_init();
1196
1197 e820_reserve_resources();
1198 e820_mark_nosave_regions(max_low_pfn);
1199
1200 x86_init.resources.reserve_resources();
1201
1202 e820_setup_gap();
1203
1204 #ifdef CONFIG_VT
1205 #if defined(CONFIG_VGA_CONSOLE)
1206 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1207 conswitchp = &vga_con;
1208 #elif defined(CONFIG_DUMMY_CONSOLE)
1209 conswitchp = &dummy_con;
1210 #endif
1211 #endif
1212 x86_init.oem.banner();
1213
1214 x86_init.timers.wallclock_init();
1215
1216 mcheck_init();
1217
1218 arch_init_ideal_nops();
1219
1220 register_refined_jiffies(CLOCK_TICK_RATE);
1221
1222 #ifdef CONFIG_EFI
1223 /* Once setup is done above, unmap the EFI memory map on
1224 * mismatched firmware/kernel archtectures since there is no
1225 * support for runtime services.
1226 */
1227 if (efi_enabled(EFI_BOOT) && !efi_is_native()) {
1228 pr_info("efi: Setup done, disabling due to 32/64-bit mismatch\n");
1229 efi_unmap_memmap();
1230 }
1231 #endif
1232 }
1233
1234 #ifdef CONFIG_X86_32
1235
1236 static struct resource video_ram_resource = {
1237 .name = "Video RAM area",
1238 .start = 0xa0000,
1239 .end = 0xbffff,
1240 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1241 };
1242
i386_reserve_resources(void)1243 void __init i386_reserve_resources(void)
1244 {
1245 request_resource(&iomem_resource, &video_ram_resource);
1246 reserve_standard_io_resources();
1247 }
1248
1249 #endif /* CONFIG_X86_32 */
1250