1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 1995 Linus Torvalds
4 *
5 * This file contains the setup_arch() code, which handles the architecture-dependent
6 * parts of early kernel initialization.
7 */
8 #include <linux/console.h>
9 #include <linux/crash_dump.h>
10 #include <linux/dma-map-ops.h>
11 #include <linux/dmi.h>
12 #include <linux/efi.h>
13 #include <linux/init_ohci1394_dma.h>
14 #include <linux/initrd.h>
15 #include <linux/iscsi_ibft.h>
16 #include <linux/memblock.h>
17 #include <linux/pci.h>
18 #include <linux/root_dev.h>
19 #include <linux/sfi.h>
20 #include <linux/hugetlb.h>
21 #include <linux/tboot.h>
22 #include <linux/usb/xhci-dbgp.h>
23 #include <linux/static_call.h>
24 #include <linux/swiotlb.h>
25
26 #include <uapi/linux/mount.h>
27
28 #include <xen/xen.h>
29
30 #include <asm/apic.h>
31 #include <asm/numa.h>
32 #include <asm/bios_ebda.h>
33 #include <asm/bugs.h>
34 #include <asm/cpu.h>
35 #include <asm/efi.h>
36 #include <asm/gart.h>
37 #include <asm/hypervisor.h>
38 #include <asm/io_apic.h>
39 #include <asm/kasan.h>
40 #include <asm/kaslr.h>
41 #include <asm/mce.h>
42 #include <asm/mtrr.h>
43 #include <asm/realmode.h>
44 #include <asm/olpc_ofw.h>
45 #include <asm/pci-direct.h>
46 #include <asm/prom.h>
47 #include <asm/proto.h>
48 #include <asm/unwind.h>
49 #include <asm/vsyscall.h>
50 #include <linux/vmalloc.h>
51
52 /*
53 * max_low_pfn_mapped: highest directly mapped pfn < 4 GB
54 * max_pfn_mapped: highest directly mapped pfn > 4 GB
55 *
56 * The direct mapping only covers E820_TYPE_RAM regions, so the ranges and gaps are
57 * represented by pfn_mapped[].
58 */
59 unsigned long max_low_pfn_mapped;
60 unsigned long max_pfn_mapped;
61
62 #ifdef CONFIG_DMI
63 RESERVE_BRK(dmi_alloc, 65536);
64 #endif
65
66
67 unsigned long _brk_start = (unsigned long)__brk_base;
68 unsigned long _brk_end = (unsigned long)__brk_base;
69
70 struct boot_params boot_params;
71
72 /*
73 * These are the four main kernel memory regions, we put them into
74 * the resource tree so that kdump tools and other debugging tools
75 * recover it:
76 */
77
78 static struct resource rodata_resource = {
79 .name = "Kernel rodata",
80 .start = 0,
81 .end = 0,
82 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
83 };
84
85 static struct resource data_resource = {
86 .name = "Kernel data",
87 .start = 0,
88 .end = 0,
89 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
90 };
91
92 static struct resource code_resource = {
93 .name = "Kernel code",
94 .start = 0,
95 .end = 0,
96 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
97 };
98
99 static struct resource bss_resource = {
100 .name = "Kernel bss",
101 .start = 0,
102 .end = 0,
103 .flags = IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
104 };
105
106
107 #ifdef CONFIG_X86_32
108 /* CPU data as detected by the assembly code in head_32.S */
109 struct cpuinfo_x86 new_cpu_data;
110
111 /* Common CPU data for all CPUs */
112 struct cpuinfo_x86 boot_cpu_data __read_mostly;
113 EXPORT_SYMBOL(boot_cpu_data);
114
115 unsigned int def_to_bigsmp;
116
117 /* For MCA, but anyone else can use it if they want */
118 unsigned int machine_id;
119 unsigned int machine_submodel_id;
120 unsigned int BIOS_revision;
121
122 struct apm_info apm_info;
123 EXPORT_SYMBOL(apm_info);
124
125 #if defined(CONFIG_X86_SPEEDSTEP_SMI) || \
126 defined(CONFIG_X86_SPEEDSTEP_SMI_MODULE)
127 struct ist_info ist_info;
128 EXPORT_SYMBOL(ist_info);
129 #else
130 struct ist_info ist_info;
131 #endif
132
133 #else
134 struct cpuinfo_x86 boot_cpu_data __read_mostly;
135 EXPORT_SYMBOL(boot_cpu_data);
136 #endif
137
138
139 #if !defined(CONFIG_X86_PAE) || defined(CONFIG_X86_64)
140 __visible unsigned long mmu_cr4_features __ro_after_init;
141 #else
142 __visible unsigned long mmu_cr4_features __ro_after_init = X86_CR4_PAE;
143 #endif
144
145 /* Boot loader ID and version as integers, for the benefit of proc_dointvec */
146 int bootloader_type, bootloader_version;
147
148 /*
149 * Setup options
150 */
151 struct screen_info screen_info;
152 EXPORT_SYMBOL(screen_info);
153 struct edid_info edid_info;
154 EXPORT_SYMBOL_GPL(edid_info);
155
156 extern int root_mountflags;
157
158 unsigned long saved_video_mode;
159
160 #define RAMDISK_IMAGE_START_MASK 0x07FF
161 #define RAMDISK_PROMPT_FLAG 0x8000
162 #define RAMDISK_LOAD_FLAG 0x4000
163
164 static char __initdata command_line[COMMAND_LINE_SIZE];
165 #ifdef CONFIG_CMDLINE_BOOL
166 static char __initdata builtin_cmdline[COMMAND_LINE_SIZE] = CONFIG_CMDLINE;
167 #endif
168
169 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
170 struct edd edd;
171 #ifdef CONFIG_EDD_MODULE
172 EXPORT_SYMBOL(edd);
173 #endif
174 /**
175 * copy_edd() - Copy the BIOS EDD information
176 * from boot_params into a safe place.
177 *
178 */
copy_edd(void)179 static inline void __init copy_edd(void)
180 {
181 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
182 sizeof(edd.mbr_signature));
183 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
184 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
185 edd.edd_info_nr = boot_params.eddbuf_entries;
186 }
187 #else
copy_edd(void)188 static inline void __init copy_edd(void)
189 {
190 }
191 #endif
192
extend_brk(size_t size,size_t align)193 void * __init extend_brk(size_t size, size_t align)
194 {
195 size_t mask = align - 1;
196 void *ret;
197
198 BUG_ON(_brk_start == 0);
199 BUG_ON(align & mask);
200
201 _brk_end = (_brk_end + mask) & ~mask;
202 BUG_ON((char *)(_brk_end + size) > __brk_limit);
203
204 ret = (void *)_brk_end;
205 _brk_end += size;
206
207 memset(ret, 0, size);
208
209 return ret;
210 }
211
212 #ifdef CONFIG_X86_32
cleanup_highmap(void)213 static void __init cleanup_highmap(void)
214 {
215 }
216 #endif
217
reserve_brk(void)218 static void __init reserve_brk(void)
219 {
220 if (_brk_end > _brk_start)
221 memblock_reserve(__pa_symbol(_brk_start),
222 _brk_end - _brk_start);
223
224 /* Mark brk area as locked down and no longer taking any
225 new allocations */
226 _brk_start = 0;
227 }
228
229 u64 relocated_ramdisk;
230
231 #ifdef CONFIG_BLK_DEV_INITRD
232
get_ramdisk_image(void)233 static u64 __init get_ramdisk_image(void)
234 {
235 u64 ramdisk_image = boot_params.hdr.ramdisk_image;
236
237 ramdisk_image |= (u64)boot_params.ext_ramdisk_image << 32;
238
239 if (ramdisk_image == 0)
240 ramdisk_image = phys_initrd_start;
241
242 return ramdisk_image;
243 }
get_ramdisk_size(void)244 static u64 __init get_ramdisk_size(void)
245 {
246 u64 ramdisk_size = boot_params.hdr.ramdisk_size;
247
248 ramdisk_size |= (u64)boot_params.ext_ramdisk_size << 32;
249
250 if (ramdisk_size == 0)
251 ramdisk_size = phys_initrd_size;
252
253 return ramdisk_size;
254 }
255
relocate_initrd(void)256 static void __init relocate_initrd(void)
257 {
258 /* Assume only end is not page aligned */
259 u64 ramdisk_image = get_ramdisk_image();
260 u64 ramdisk_size = get_ramdisk_size();
261 u64 area_size = PAGE_ALIGN(ramdisk_size);
262
263 /* We need to move the initrd down into directly mapped mem */
264 relocated_ramdisk = memblock_phys_alloc_range(area_size, PAGE_SIZE, 0,
265 PFN_PHYS(max_pfn_mapped));
266 if (!relocated_ramdisk)
267 panic("Cannot find place for new RAMDISK of size %lld\n",
268 ramdisk_size);
269
270 initrd_start = relocated_ramdisk + PAGE_OFFSET;
271 initrd_end = initrd_start + ramdisk_size;
272 printk(KERN_INFO "Allocated new RAMDISK: [mem %#010llx-%#010llx]\n",
273 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
274
275 copy_from_early_mem((void *)initrd_start, ramdisk_image, ramdisk_size);
276
277 printk(KERN_INFO "Move RAMDISK from [mem %#010llx-%#010llx] to"
278 " [mem %#010llx-%#010llx]\n",
279 ramdisk_image, ramdisk_image + ramdisk_size - 1,
280 relocated_ramdisk, relocated_ramdisk + ramdisk_size - 1);
281 }
282
early_reserve_initrd(void)283 static void __init early_reserve_initrd(void)
284 {
285 /* Assume only end is not page aligned */
286 u64 ramdisk_image = get_ramdisk_image();
287 u64 ramdisk_size = get_ramdisk_size();
288 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
289
290 if (!boot_params.hdr.type_of_loader ||
291 !ramdisk_image || !ramdisk_size)
292 return; /* No initrd provided by bootloader */
293
294 memblock_reserve(ramdisk_image, ramdisk_end - ramdisk_image);
295 }
296
reserve_initrd(void)297 static void __init reserve_initrd(void)
298 {
299 /* Assume only end is not page aligned */
300 u64 ramdisk_image = get_ramdisk_image();
301 u64 ramdisk_size = get_ramdisk_size();
302 u64 ramdisk_end = PAGE_ALIGN(ramdisk_image + ramdisk_size);
303
304 if (!boot_params.hdr.type_of_loader ||
305 !ramdisk_image || !ramdisk_size)
306 return; /* No initrd provided by bootloader */
307
308 initrd_start = 0;
309
310 printk(KERN_INFO "RAMDISK: [mem %#010llx-%#010llx]\n", ramdisk_image,
311 ramdisk_end - 1);
312
313 if (pfn_range_is_mapped(PFN_DOWN(ramdisk_image),
314 PFN_DOWN(ramdisk_end))) {
315 /* All are mapped, easy case */
316 initrd_start = ramdisk_image + PAGE_OFFSET;
317 initrd_end = initrd_start + ramdisk_size;
318 return;
319 }
320
321 relocate_initrd();
322
323 memblock_free(ramdisk_image, ramdisk_end - ramdisk_image);
324 }
325
326 #else
early_reserve_initrd(void)327 static void __init early_reserve_initrd(void)
328 {
329 }
reserve_initrd(void)330 static void __init reserve_initrd(void)
331 {
332 }
333 #endif /* CONFIG_BLK_DEV_INITRD */
334
parse_setup_data(void)335 static void __init parse_setup_data(void)
336 {
337 struct setup_data *data;
338 u64 pa_data, pa_next;
339
340 pa_data = boot_params.hdr.setup_data;
341 while (pa_data) {
342 u32 data_len, data_type;
343
344 data = early_memremap(pa_data, sizeof(*data));
345 data_len = data->len + sizeof(struct setup_data);
346 data_type = data->type;
347 pa_next = data->next;
348 early_memunmap(data, sizeof(*data));
349
350 switch (data_type) {
351 case SETUP_E820_EXT:
352 e820__memory_setup_extended(pa_data, data_len);
353 break;
354 case SETUP_DTB:
355 add_dtb(pa_data);
356 break;
357 case SETUP_EFI:
358 parse_efi_setup(pa_data, data_len);
359 break;
360 default:
361 break;
362 }
363 pa_data = pa_next;
364 }
365 }
366
memblock_x86_reserve_range_setup_data(void)367 static void __init memblock_x86_reserve_range_setup_data(void)
368 {
369 struct setup_indirect *indirect;
370 struct setup_data *data;
371 u64 pa_data, pa_next;
372 u32 len;
373
374 pa_data = boot_params.hdr.setup_data;
375 while (pa_data) {
376 data = early_memremap(pa_data, sizeof(*data));
377 if (!data) {
378 pr_warn("setup: failed to memremap setup_data entry\n");
379 return;
380 }
381
382 len = sizeof(*data);
383 pa_next = data->next;
384
385 memblock_reserve(pa_data, sizeof(*data) + data->len);
386
387 if (data->type == SETUP_INDIRECT) {
388 len += data->len;
389 early_memunmap(data, sizeof(*data));
390 data = early_memremap(pa_data, len);
391 if (!data) {
392 pr_warn("setup: failed to memremap indirect setup_data\n");
393 return;
394 }
395
396 indirect = (struct setup_indirect *)data->data;
397
398 if (indirect->type != SETUP_INDIRECT)
399 memblock_reserve(indirect->addr, indirect->len);
400 }
401
402 pa_data = pa_next;
403 early_memunmap(data, len);
404 }
405 }
406
407 /*
408 * --------- Crashkernel reservation ------------------------------
409 */
410
411 #ifdef CONFIG_KEXEC_CORE
412
413 /* 16M alignment for crash kernel regions */
414 #define CRASH_ALIGN SZ_16M
415
416 /*
417 * Keep the crash kernel below this limit.
418 *
419 * Earlier 32-bits kernels would limit the kernel to the low 512 MB range
420 * due to mapping restrictions.
421 *
422 * 64-bit kdump kernels need to be restricted to be under 64 TB, which is
423 * the upper limit of system RAM in 4-level paging mode. Since the kdump
424 * jump could be from 5-level paging to 4-level paging, the jump will fail if
425 * the kernel is put above 64 TB, and during the 1st kernel bootup there's
426 * no good way to detect the paging mode of the target kernel which will be
427 * loaded for dumping.
428 */
429 #ifdef CONFIG_X86_32
430 # define CRASH_ADDR_LOW_MAX SZ_512M
431 # define CRASH_ADDR_HIGH_MAX SZ_512M
432 #else
433 # define CRASH_ADDR_LOW_MAX SZ_4G
434 # define CRASH_ADDR_HIGH_MAX SZ_64T
435 #endif
436
reserve_crashkernel_low(void)437 static int __init reserve_crashkernel_low(void)
438 {
439 #ifdef CONFIG_X86_64
440 unsigned long long base, low_base = 0, low_size = 0;
441 unsigned long low_mem_limit;
442 int ret;
443
444 low_mem_limit = min(memblock_phys_mem_size(), CRASH_ADDR_LOW_MAX);
445
446 /* crashkernel=Y,low */
447 ret = parse_crashkernel_low(boot_command_line, low_mem_limit, &low_size, &base);
448 if (ret) {
449 /*
450 * two parts from kernel/dma/swiotlb.c:
451 * -swiotlb size: user-specified with swiotlb= or default.
452 *
453 * -swiotlb overflow buffer: now hardcoded to 32k. We round it
454 * to 8M for other buffers that may need to stay low too. Also
455 * make sure we allocate enough extra low memory so that we
456 * don't run out of DMA buffers for 32-bit devices.
457 */
458 low_size = max(swiotlb_size_or_default() + (8UL << 20), 256UL << 20);
459 } else {
460 /* passed with crashkernel=0,low ? */
461 if (!low_size)
462 return 0;
463 }
464
465 low_base = memblock_phys_alloc_range(low_size, CRASH_ALIGN, 0, CRASH_ADDR_LOW_MAX);
466 if (!low_base) {
467 pr_err("Cannot reserve %ldMB crashkernel low memory, please try smaller size.\n",
468 (unsigned long)(low_size >> 20));
469 return -ENOMEM;
470 }
471
472 pr_info("Reserving %ldMB of low memory at %ldMB for crashkernel (low RAM limit: %ldMB)\n",
473 (unsigned long)(low_size >> 20),
474 (unsigned long)(low_base >> 20),
475 (unsigned long)(low_mem_limit >> 20));
476
477 crashk_low_res.start = low_base;
478 crashk_low_res.end = low_base + low_size - 1;
479 insert_resource(&iomem_resource, &crashk_low_res);
480 #endif
481 return 0;
482 }
483
reserve_crashkernel(void)484 static void __init reserve_crashkernel(void)
485 {
486 unsigned long long crash_size, crash_base, total_mem;
487 bool high = false;
488 int ret;
489
490 total_mem = memblock_phys_mem_size();
491
492 /* crashkernel=XM */
493 ret = parse_crashkernel(boot_command_line, total_mem, &crash_size, &crash_base);
494 if (ret != 0 || crash_size <= 0) {
495 /* crashkernel=X,high */
496 ret = parse_crashkernel_high(boot_command_line, total_mem,
497 &crash_size, &crash_base);
498 if (ret != 0 || crash_size <= 0)
499 return;
500 high = true;
501 }
502
503 if (xen_pv_domain()) {
504 pr_info("Ignoring crashkernel for a Xen PV domain\n");
505 return;
506 }
507
508 /* 0 means: find the address automatically */
509 if (!crash_base) {
510 /*
511 * Set CRASH_ADDR_LOW_MAX upper bound for crash memory,
512 * crashkernel=x,high reserves memory over 4G, also allocates
513 * 256M extra low memory for DMA buffers and swiotlb.
514 * But the extra memory is not required for all machines.
515 * So try low memory first and fall back to high memory
516 * unless "crashkernel=size[KMG],high" is specified.
517 */
518 if (!high)
519 crash_base = memblock_phys_alloc_range(crash_size,
520 CRASH_ALIGN, CRASH_ALIGN,
521 CRASH_ADDR_LOW_MAX);
522 if (!crash_base)
523 crash_base = memblock_phys_alloc_range(crash_size,
524 CRASH_ALIGN, CRASH_ALIGN,
525 CRASH_ADDR_HIGH_MAX);
526 if (!crash_base) {
527 pr_info("crashkernel reservation failed - No suitable area found.\n");
528 return;
529 }
530 } else {
531 unsigned long long start;
532
533 start = memblock_phys_alloc_range(crash_size, SZ_1M, crash_base,
534 crash_base + crash_size);
535 if (start != crash_base) {
536 pr_info("crashkernel reservation failed - memory is in use.\n");
537 return;
538 }
539 }
540
541 if (crash_base >= (1ULL << 32) && reserve_crashkernel_low()) {
542 memblock_free(crash_base, crash_size);
543 return;
544 }
545
546 pr_info("Reserving %ldMB of memory at %ldMB for crashkernel (System RAM: %ldMB)\n",
547 (unsigned long)(crash_size >> 20),
548 (unsigned long)(crash_base >> 20),
549 (unsigned long)(total_mem >> 20));
550
551 crashk_res.start = crash_base;
552 crashk_res.end = crash_base + crash_size - 1;
553 insert_resource(&iomem_resource, &crashk_res);
554 }
555 #else
reserve_crashkernel(void)556 static void __init reserve_crashkernel(void)
557 {
558 }
559 #endif
560
561 static struct resource standard_io_resources[] = {
562 { .name = "dma1", .start = 0x00, .end = 0x1f,
563 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
564 { .name = "pic1", .start = 0x20, .end = 0x21,
565 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
566 { .name = "timer0", .start = 0x40, .end = 0x43,
567 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
568 { .name = "timer1", .start = 0x50, .end = 0x53,
569 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
570 { .name = "keyboard", .start = 0x60, .end = 0x60,
571 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
572 { .name = "keyboard", .start = 0x64, .end = 0x64,
573 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
574 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
575 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
576 { .name = "pic2", .start = 0xa0, .end = 0xa1,
577 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
578 { .name = "dma2", .start = 0xc0, .end = 0xdf,
579 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
580 { .name = "fpu", .start = 0xf0, .end = 0xff,
581 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
582 };
583
reserve_standard_io_resources(void)584 void __init reserve_standard_io_resources(void)
585 {
586 int i;
587
588 /* request I/O space for devices used on all i[345]86 PCs */
589 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
590 request_resource(&ioport_resource, &standard_io_resources[i]);
591
592 }
593
reserve_ibft_region(void)594 static __init void reserve_ibft_region(void)
595 {
596 unsigned long addr, size = 0;
597
598 addr = find_ibft_region(&size);
599
600 if (size)
601 memblock_reserve(addr, size);
602 }
603
snb_gfx_workaround_needed(void)604 static bool __init snb_gfx_workaround_needed(void)
605 {
606 #ifdef CONFIG_PCI
607 int i;
608 u16 vendor, devid;
609 static const __initconst u16 snb_ids[] = {
610 0x0102,
611 0x0112,
612 0x0122,
613 0x0106,
614 0x0116,
615 0x0126,
616 0x010a,
617 };
618
619 /* Assume no if something weird is going on with PCI */
620 if (!early_pci_allowed())
621 return false;
622
623 vendor = read_pci_config_16(0, 2, 0, PCI_VENDOR_ID);
624 if (vendor != 0x8086)
625 return false;
626
627 devid = read_pci_config_16(0, 2, 0, PCI_DEVICE_ID);
628 for (i = 0; i < ARRAY_SIZE(snb_ids); i++)
629 if (devid == snb_ids[i])
630 return true;
631 #endif
632
633 return false;
634 }
635
636 /*
637 * Sandy Bridge graphics has trouble with certain ranges, exclude
638 * them from allocation.
639 */
trim_snb_memory(void)640 static void __init trim_snb_memory(void)
641 {
642 static const __initconst unsigned long bad_pages[] = {
643 0x20050000,
644 0x20110000,
645 0x20130000,
646 0x20138000,
647 0x40004000,
648 };
649 int i;
650
651 if (!snb_gfx_workaround_needed())
652 return;
653
654 printk(KERN_DEBUG "reserving inaccessible SNB gfx pages\n");
655
656 /*
657 * Reserve all memory below the 1 MB mark that has not
658 * already been reserved.
659 */
660 memblock_reserve(0, 1<<20);
661
662 for (i = 0; i < ARRAY_SIZE(bad_pages); i++) {
663 if (memblock_reserve(bad_pages[i], PAGE_SIZE))
664 printk(KERN_WARNING "failed to reserve 0x%08lx\n",
665 bad_pages[i]);
666 }
667 }
668
669 /*
670 * Here we put platform-specific memory range workarounds, i.e.
671 * memory known to be corrupt or otherwise in need to be reserved on
672 * specific platforms.
673 *
674 * If this gets used more widely it could use a real dispatch mechanism.
675 */
trim_platform_memory_ranges(void)676 static void __init trim_platform_memory_ranges(void)
677 {
678 trim_snb_memory();
679 }
680
trim_bios_range(void)681 static void __init trim_bios_range(void)
682 {
683 /*
684 * A special case is the first 4Kb of memory;
685 * This is a BIOS owned area, not kernel ram, but generally
686 * not listed as such in the E820 table.
687 *
688 * This typically reserves additional memory (64KiB by default)
689 * since some BIOSes are known to corrupt low memory. See the
690 * Kconfig help text for X86_RESERVE_LOW.
691 */
692 e820__range_update(0, PAGE_SIZE, E820_TYPE_RAM, E820_TYPE_RESERVED);
693
694 /*
695 * special case: Some BIOSes report the PC BIOS
696 * area (640Kb -> 1Mb) as RAM even though it is not.
697 * take them out.
698 */
699 e820__range_remove(BIOS_BEGIN, BIOS_END - BIOS_BEGIN, E820_TYPE_RAM, 1);
700
701 e820__update_table(e820_table);
702 }
703
704 /* called before trim_bios_range() to spare extra sanitize */
e820_add_kernel_range(void)705 static void __init e820_add_kernel_range(void)
706 {
707 u64 start = __pa_symbol(_text);
708 u64 size = __pa_symbol(_end) - start;
709
710 /*
711 * Complain if .text .data and .bss are not marked as E820_TYPE_RAM and
712 * attempt to fix it by adding the range. We may have a confused BIOS,
713 * or the user may have used memmap=exactmap or memmap=xxM$yyM to
714 * exclude kernel range. If we really are running on top non-RAM,
715 * we will crash later anyways.
716 */
717 if (e820__mapped_all(start, start + size, E820_TYPE_RAM))
718 return;
719
720 pr_warn(".text .data .bss are not marked as E820_TYPE_RAM!\n");
721 e820__range_remove(start, size, E820_TYPE_RAM, 0);
722 e820__range_add(start, size, E820_TYPE_RAM);
723 }
724
725 static unsigned reserve_low = CONFIG_X86_RESERVE_LOW << 10;
726
parse_reservelow(char * p)727 static int __init parse_reservelow(char *p)
728 {
729 unsigned long long size;
730
731 if (!p)
732 return -EINVAL;
733
734 size = memparse(p, &p);
735
736 if (size < 4096)
737 size = 4096;
738
739 if (size > 640*1024)
740 size = 640*1024;
741
742 reserve_low = size;
743
744 return 0;
745 }
746
747 early_param("reservelow", parse_reservelow);
748
trim_low_memory_range(void)749 static void __init trim_low_memory_range(void)
750 {
751 memblock_reserve(0, ALIGN(reserve_low, PAGE_SIZE));
752 }
753
754 /*
755 * Dump out kernel offset information on panic.
756 */
757 static int
dump_kernel_offset(struct notifier_block * self,unsigned long v,void * p)758 dump_kernel_offset(struct notifier_block *self, unsigned long v, void *p)
759 {
760 if (kaslr_enabled()) {
761 pr_emerg("Kernel Offset: 0x%lx from 0x%lx (relocation range: 0x%lx-0x%lx)\n",
762 kaslr_offset(),
763 __START_KERNEL,
764 __START_KERNEL_map,
765 MODULES_VADDR-1);
766 } else {
767 pr_emerg("Kernel Offset: disabled\n");
768 }
769
770 return 0;
771 }
772
773 /*
774 * Determine if we were loaded by an EFI loader. If so, then we have also been
775 * passed the efi memmap, systab, etc., so we should use these data structures
776 * for initialization. Note, the efi init code path is determined by the
777 * global efi_enabled. This allows the same kernel image to be used on existing
778 * systems (with a traditional BIOS) as well as on EFI systems.
779 */
780 /*
781 * setup_arch - architecture-specific boot-time initializations
782 *
783 * Note: On x86_64, fixmaps are ready for use even before this is called.
784 */
785
setup_arch(char ** cmdline_p)786 void __init setup_arch(char **cmdline_p)
787 {
788 /*
789 * Reserve the memory occupied by the kernel between _text and
790 * __end_of_kernel_reserve symbols. Any kernel sections after the
791 * __end_of_kernel_reserve symbol must be explicitly reserved with a
792 * separate memblock_reserve() or they will be discarded.
793 */
794 memblock_reserve(__pa_symbol(_text),
795 (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
796
797 /*
798 * Make sure page 0 is always reserved because on systems with
799 * L1TF its contents can be leaked to user processes.
800 */
801 memblock_reserve(0, PAGE_SIZE);
802
803 early_reserve_initrd();
804
805 /*
806 * At this point everything still needed from the boot loader
807 * or BIOS or kernel text should be early reserved or marked not
808 * RAM in e820. All other memory is free game.
809 */
810
811 #ifdef CONFIG_X86_32
812 memcpy(&boot_cpu_data, &new_cpu_data, sizeof(new_cpu_data));
813
814 /*
815 * copy kernel address range established so far and switch
816 * to the proper swapper page table
817 */
818 clone_pgd_range(swapper_pg_dir + KERNEL_PGD_BOUNDARY,
819 initial_page_table + KERNEL_PGD_BOUNDARY,
820 KERNEL_PGD_PTRS);
821
822 load_cr3(swapper_pg_dir);
823 /*
824 * Note: Quark X1000 CPUs advertise PGE incorrectly and require
825 * a cr3 based tlb flush, so the following __flush_tlb_all()
826 * will not flush anything because the CPU quirk which clears
827 * X86_FEATURE_PGE has not been invoked yet. Though due to the
828 * load_cr3() above the TLB has been flushed already. The
829 * quirk is invoked before subsequent calls to __flush_tlb_all()
830 * so proper operation is guaranteed.
831 */
832 __flush_tlb_all();
833 #else
834 printk(KERN_INFO "Command line: %s\n", boot_command_line);
835 boot_cpu_data.x86_phys_bits = MAX_PHYSMEM_BITS;
836 #endif
837
838 /*
839 * If we have OLPC OFW, we might end up relocating the fixmap due to
840 * reserve_top(), so do this before touching the ioremap area.
841 */
842 olpc_ofw_detect();
843
844 idt_setup_early_traps();
845 early_cpu_init();
846 arch_init_ideal_nops();
847 jump_label_init();
848 static_call_init();
849 early_ioremap_init();
850
851 setup_olpc_ofw_pgd();
852
853 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
854 screen_info = boot_params.screen_info;
855 edid_info = boot_params.edid_info;
856 #ifdef CONFIG_X86_32
857 apm_info.bios = boot_params.apm_bios_info;
858 ist_info = boot_params.ist_info;
859 #endif
860 saved_video_mode = boot_params.hdr.vid_mode;
861 bootloader_type = boot_params.hdr.type_of_loader;
862 if ((bootloader_type >> 4) == 0xe) {
863 bootloader_type &= 0xf;
864 bootloader_type |= (boot_params.hdr.ext_loader_type+0x10) << 4;
865 }
866 bootloader_version = bootloader_type & 0xf;
867 bootloader_version |= boot_params.hdr.ext_loader_ver << 4;
868
869 #ifdef CONFIG_BLK_DEV_RAM
870 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
871 #endif
872 #ifdef CONFIG_EFI
873 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
874 EFI32_LOADER_SIGNATURE, 4)) {
875 set_bit(EFI_BOOT, &efi.flags);
876 } else if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
877 EFI64_LOADER_SIGNATURE, 4)) {
878 set_bit(EFI_BOOT, &efi.flags);
879 set_bit(EFI_64BIT, &efi.flags);
880 }
881 #endif
882
883 x86_init.oem.arch_setup();
884
885 iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
886 e820__memory_setup();
887 parse_setup_data();
888
889 copy_edd();
890
891 if (!boot_params.hdr.root_flags)
892 root_mountflags &= ~MS_RDONLY;
893 init_mm.start_code = (unsigned long) _text;
894 init_mm.end_code = (unsigned long) _etext;
895 init_mm.end_data = (unsigned long) _edata;
896 init_mm.brk = _brk_end;
897
898 code_resource.start = __pa_symbol(_text);
899 code_resource.end = __pa_symbol(_etext)-1;
900 rodata_resource.start = __pa_symbol(__start_rodata);
901 rodata_resource.end = __pa_symbol(__end_rodata)-1;
902 data_resource.start = __pa_symbol(_sdata);
903 data_resource.end = __pa_symbol(_edata)-1;
904 bss_resource.start = __pa_symbol(__bss_start);
905 bss_resource.end = __pa_symbol(__bss_stop)-1;
906
907 #ifdef CONFIG_CMDLINE_BOOL
908 #ifdef CONFIG_CMDLINE_OVERRIDE
909 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
910 #else
911 if (builtin_cmdline[0]) {
912 /* append boot loader cmdline to builtin */
913 strlcat(builtin_cmdline, " ", COMMAND_LINE_SIZE);
914 strlcat(builtin_cmdline, boot_command_line, COMMAND_LINE_SIZE);
915 strlcpy(boot_command_line, builtin_cmdline, COMMAND_LINE_SIZE);
916 }
917 #endif
918 #endif
919
920 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
921 *cmdline_p = command_line;
922
923 /*
924 * x86_configure_nx() is called before parse_early_param() to detect
925 * whether hardware doesn't support NX (so that the early EHCI debug
926 * console setup can safely call set_fixmap()). It may then be called
927 * again from within noexec_setup() during parsing early parameters
928 * to honor the respective command line option.
929 */
930 x86_configure_nx();
931
932 parse_early_param();
933
934 if (efi_enabled(EFI_BOOT))
935 efi_memblock_x86_reserve_range();
936 #ifdef CONFIG_MEMORY_HOTPLUG
937 /*
938 * Memory used by the kernel cannot be hot-removed because Linux
939 * cannot migrate the kernel pages. When memory hotplug is
940 * enabled, we should prevent memblock from allocating memory
941 * for the kernel.
942 *
943 * ACPI SRAT records all hotpluggable memory ranges. But before
944 * SRAT is parsed, we don't know about it.
945 *
946 * The kernel image is loaded into memory at very early time. We
947 * cannot prevent this anyway. So on NUMA system, we set any
948 * node the kernel resides in as un-hotpluggable.
949 *
950 * Since on modern servers, one node could have double-digit
951 * gigabytes memory, we can assume the memory around the kernel
952 * image is also un-hotpluggable. So before SRAT is parsed, just
953 * allocate memory near the kernel image to try the best to keep
954 * the kernel away from hotpluggable memory.
955 */
956 if (movable_node_is_enabled())
957 memblock_set_bottom_up(true);
958 #endif
959
960 x86_report_nx();
961
962 /* after early param, so could get panic from serial */
963 memblock_x86_reserve_range_setup_data();
964
965 if (acpi_mps_check()) {
966 #ifdef CONFIG_X86_LOCAL_APIC
967 disable_apic = 1;
968 #endif
969 setup_clear_cpu_cap(X86_FEATURE_APIC);
970 }
971
972 e820__reserve_setup_data();
973 e820__finish_early_params();
974
975 if (efi_enabled(EFI_BOOT))
976 efi_init();
977
978 dmi_setup();
979
980 /*
981 * VMware detection requires dmi to be available, so this
982 * needs to be done after dmi_setup(), for the boot CPU.
983 */
984 init_hypervisor_platform();
985
986 tsc_early_init();
987 x86_init.resources.probe_roms();
988
989 /* after parse_early_param, so could debug it */
990 insert_resource(&iomem_resource, &code_resource);
991 insert_resource(&iomem_resource, &rodata_resource);
992 insert_resource(&iomem_resource, &data_resource);
993 insert_resource(&iomem_resource, &bss_resource);
994
995 e820_add_kernel_range();
996 trim_bios_range();
997 #ifdef CONFIG_X86_32
998 if (ppro_with_ram_bug()) {
999 e820__range_update(0x70000000ULL, 0x40000ULL, E820_TYPE_RAM,
1000 E820_TYPE_RESERVED);
1001 e820__update_table(e820_table);
1002 printk(KERN_INFO "fixed physical RAM map:\n");
1003 e820__print_table("bad_ppro");
1004 }
1005 #else
1006 early_gart_iommu_check();
1007 #endif
1008
1009 /*
1010 * partially used pages are not usable - thus
1011 * we are rounding upwards:
1012 */
1013 max_pfn = e820__end_of_ram_pfn();
1014
1015 /* update e820 for memory not covered by WB MTRRs */
1016 mtrr_bp_init();
1017 if (mtrr_trim_uncached_memory(max_pfn))
1018 max_pfn = e820__end_of_ram_pfn();
1019
1020 max_possible_pfn = max_pfn;
1021
1022 /*
1023 * This call is required when the CPU does not support PAT. If
1024 * mtrr_bp_init() invoked it already via pat_init() the call has no
1025 * effect.
1026 */
1027 init_cache_modes();
1028
1029 /*
1030 * Define random base addresses for memory sections after max_pfn is
1031 * defined and before each memory section base is used.
1032 */
1033 kernel_randomize_memory();
1034
1035 #ifdef CONFIG_X86_32
1036 /* max_low_pfn get updated here */
1037 find_low_pfn_range();
1038 #else
1039 check_x2apic();
1040
1041 /* How many end-of-memory variables you have, grandma! */
1042 /* need this before calling reserve_initrd */
1043 if (max_pfn > (1UL<<(32 - PAGE_SHIFT)))
1044 max_low_pfn = e820__end_of_low_ram_pfn();
1045 else
1046 max_low_pfn = max_pfn;
1047
1048 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
1049 #endif
1050
1051 /*
1052 * Find and reserve possible boot-time SMP configuration:
1053 */
1054 find_smp_config();
1055
1056 reserve_ibft_region();
1057
1058 early_alloc_pgt_buf();
1059
1060 /*
1061 * Need to conclude brk, before e820__memblock_setup()
1062 * it could use memblock_find_in_range, could overlap with
1063 * brk area.
1064 */
1065 reserve_brk();
1066
1067 cleanup_highmap();
1068
1069 memblock_set_current_limit(ISA_END_ADDRESS);
1070 e820__memblock_setup();
1071
1072 reserve_bios_regions();
1073
1074 efi_fake_memmap();
1075 efi_find_mirror();
1076 efi_esrt_init();
1077 efi_mokvar_table_init();
1078
1079 /*
1080 * The EFI specification says that boot service code won't be
1081 * called after ExitBootServices(). This is, in fact, a lie.
1082 */
1083 efi_reserve_boot_services();
1084
1085 /* preallocate 4k for mptable mpc */
1086 e820__memblock_alloc_reserved_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 idt_setup_early_pf();
1105
1106 /*
1107 * Update mmu_cr4_features (and, indirectly, trampoline_cr4_features)
1108 * with the current CR4 value. This may not be necessary, but
1109 * auditing all the early-boot CR4 manipulation would be needed to
1110 * rule it out.
1111 *
1112 * Mask off features that don't work outside long mode (just
1113 * PCIDE for now).
1114 */
1115 mmu_cr4_features = __read_cr4() & ~X86_CR4_PCIDE;
1116
1117 memblock_set_current_limit(get_max_mapped());
1118
1119 /*
1120 * NOTE: On x86-32, only from this point on, fixmaps are ready for use.
1121 */
1122
1123 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
1124 if (init_ohci1394_dma_early)
1125 init_ohci1394_dma_on_all_controllers();
1126 #endif
1127 /* Allocate bigger log buffer */
1128 setup_log_buf(1);
1129
1130 if (efi_enabled(EFI_BOOT)) {
1131 switch (boot_params.secure_boot) {
1132 case efi_secureboot_mode_disabled:
1133 pr_info("Secure boot disabled\n");
1134 break;
1135 case efi_secureboot_mode_enabled:
1136 pr_info("Secure boot enabled\n");
1137 break;
1138 default:
1139 pr_info("Secure boot could not be determined\n");
1140 break;
1141 }
1142 }
1143
1144 reserve_initrd();
1145
1146 acpi_table_upgrade();
1147 /* Look for ACPI tables and reserve memory occupied by them. */
1148 acpi_boot_table_init();
1149
1150 vsmp_init();
1151
1152 io_delay_init();
1153
1154 early_platform_quirks();
1155
1156 early_acpi_boot_init();
1157
1158 initmem_init();
1159 dma_contiguous_reserve(max_pfn_mapped << PAGE_SHIFT);
1160
1161 if (boot_cpu_has(X86_FEATURE_GBPAGES))
1162 hugetlb_cma_reserve(PUD_SHIFT - PAGE_SHIFT);
1163
1164 /*
1165 * Reserve memory for crash kernel after SRAT is parsed so that it
1166 * won't consume hotpluggable memory.
1167 */
1168 reserve_crashkernel();
1169
1170 memblock_find_dma_reserve();
1171
1172 if (!early_xdbc_setup_hardware())
1173 early_xdbc_register_console();
1174
1175 x86_init.paging.pagetable_init();
1176
1177 kasan_init();
1178
1179 /*
1180 * Sync back kernel address range.
1181 *
1182 * FIXME: Can the later sync in setup_cpu_entry_areas() replace
1183 * this call?
1184 */
1185 sync_initial_page_table();
1186
1187 tboot_probe();
1188
1189 map_vsyscall();
1190
1191 generic_apic_probe();
1192
1193 early_quirks();
1194
1195 /*
1196 * Read APIC and some other early information from ACPI tables.
1197 */
1198 acpi_boot_init();
1199 sfi_init();
1200 x86_dtb_init();
1201
1202 /*
1203 * get boot-time SMP configuration:
1204 */
1205 get_smp_config();
1206
1207 /*
1208 * Systems w/o ACPI and mptables might not have it mapped the local
1209 * APIC yet, but prefill_possible_map() might need to access it.
1210 */
1211 init_apic_mappings();
1212
1213 prefill_possible_map();
1214
1215 init_cpu_to_node();
1216 init_gi_nodes();
1217
1218 io_apic_init_mappings();
1219
1220 x86_init.hyper.guest_late_init();
1221
1222 e820__reserve_resources();
1223 e820__register_nosave_regions(max_pfn);
1224
1225 x86_init.resources.reserve_resources();
1226
1227 e820__setup_pci_gap();
1228
1229 #ifdef CONFIG_VT
1230 #if defined(CONFIG_VGA_CONSOLE)
1231 if (!efi_enabled(EFI_BOOT) || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
1232 conswitchp = &vga_con;
1233 #endif
1234 #endif
1235 x86_init.oem.banner();
1236
1237 x86_init.timers.wallclock_init();
1238
1239 mcheck_init();
1240
1241 register_refined_jiffies(CLOCK_TICK_RATE);
1242
1243 #ifdef CONFIG_EFI
1244 if (efi_enabled(EFI_BOOT))
1245 efi_apply_memmap_quirks();
1246 #endif
1247
1248 unwind_init();
1249 }
1250
1251 #ifdef CONFIG_X86_32
1252
1253 static struct resource video_ram_resource = {
1254 .name = "Video RAM area",
1255 .start = 0xa0000,
1256 .end = 0xbffff,
1257 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
1258 };
1259
i386_reserve_resources(void)1260 void __init i386_reserve_resources(void)
1261 {
1262 request_resource(&iomem_resource, &video_ram_resource);
1263 reserve_standard_io_resources();
1264 }
1265
1266 #endif /* CONFIG_X86_32 */
1267
1268 static struct notifier_block kernel_offset_notifier = {
1269 .notifier_call = dump_kernel_offset
1270 };
1271
register_kernel_offset_dumper(void)1272 static int __init register_kernel_offset_dumper(void)
1273 {
1274 atomic_notifier_chain_register(&panic_notifier_list,
1275 &kernel_offset_notifier);
1276 return 0;
1277 }
1278 __initcall(register_kernel_offset_dumper);
1279