Loading...
1/*
2 * handle transition of Linux booting another kernel
3 * Copyright (C) 2002-2005 Eric Biederman <ebiederm@xmission.com>
4 *
5 * This source code is licensed under the GNU General Public License,
6 * Version 2. See the file COPYING for more details.
7 */
8
9#define pr_fmt(fmt) "kexec: " fmt
10
11#include <linux/mm.h>
12#include <linux/kexec.h>
13#include <linux/string.h>
14#include <linux/gfp.h>
15#include <linux/reboot.h>
16#include <linux/numa.h>
17#include <linux/ftrace.h>
18#include <linux/io.h>
19#include <linux/suspend.h>
20#include <linux/vmalloc.h>
21
22#include <asm/init.h>
23#include <asm/pgtable.h>
24#include <asm/tlbflush.h>
25#include <asm/mmu_context.h>
26#include <asm/io_apic.h>
27#include <asm/debugreg.h>
28#include <asm/kexec-bzimage64.h>
29#include <asm/setup.h>
30#include <asm/set_memory.h>
31
32#ifdef CONFIG_KEXEC_FILE
33const struct kexec_file_ops * const kexec_file_loaders[] = {
34 &kexec_bzImage64_ops,
35 NULL
36};
37#endif
38
39static void free_transition_pgtable(struct kimage *image)
40{
41 free_page((unsigned long)image->arch.p4d);
42 image->arch.p4d = NULL;
43 free_page((unsigned long)image->arch.pud);
44 image->arch.pud = NULL;
45 free_page((unsigned long)image->arch.pmd);
46 image->arch.pmd = NULL;
47 free_page((unsigned long)image->arch.pte);
48 image->arch.pte = NULL;
49}
50
51static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
52{
53 p4d_t *p4d;
54 pud_t *pud;
55 pmd_t *pmd;
56 pte_t *pte;
57 unsigned long vaddr, paddr;
58 int result = -ENOMEM;
59
60 vaddr = (unsigned long)relocate_kernel;
61 paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
62 pgd += pgd_index(vaddr);
63 if (!pgd_present(*pgd)) {
64 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
65 if (!p4d)
66 goto err;
67 image->arch.p4d = p4d;
68 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
69 }
70 p4d = p4d_offset(pgd, vaddr);
71 if (!p4d_present(*p4d)) {
72 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
73 if (!pud)
74 goto err;
75 image->arch.pud = pud;
76 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
77 }
78 pud = pud_offset(p4d, vaddr);
79 if (!pud_present(*pud)) {
80 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
81 if (!pmd)
82 goto err;
83 image->arch.pmd = pmd;
84 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
85 }
86 pmd = pmd_offset(pud, vaddr);
87 if (!pmd_present(*pmd)) {
88 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
89 if (!pte)
90 goto err;
91 image->arch.pte = pte;
92 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
93 }
94 pte = pte_offset_kernel(pmd, vaddr);
95 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC_NOENC));
96 return 0;
97err:
98 return result;
99}
100
101static void *alloc_pgt_page(void *data)
102{
103 struct kimage *image = (struct kimage *)data;
104 struct page *page;
105 void *p = NULL;
106
107 page = kimage_alloc_control_pages(image, 0);
108 if (page) {
109 p = page_address(page);
110 clear_page(p);
111 }
112
113 return p;
114}
115
116static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
117{
118 struct x86_mapping_info info = {
119 .alloc_pgt_page = alloc_pgt_page,
120 .context = image,
121 .page_flag = __PAGE_KERNEL_LARGE_EXEC,
122 .kernpg_flag = _KERNPG_TABLE_NOENC,
123 };
124 unsigned long mstart, mend;
125 pgd_t *level4p;
126 int result;
127 int i;
128
129 level4p = (pgd_t *)__va(start_pgtable);
130 clear_page(level4p);
131
132 if (direct_gbpages)
133 info.direct_gbpages = true;
134
135 for (i = 0; i < nr_pfn_mapped; i++) {
136 mstart = pfn_mapped[i].start << PAGE_SHIFT;
137 mend = pfn_mapped[i].end << PAGE_SHIFT;
138
139 result = kernel_ident_mapping_init(&info,
140 level4p, mstart, mend);
141 if (result)
142 return result;
143 }
144
145 /*
146 * segments's mem ranges could be outside 0 ~ max_pfn,
147 * for example when jump back to original kernel from kexeced kernel.
148 * or first kernel is booted with user mem map, and second kernel
149 * could be loaded out of that range.
150 */
151 for (i = 0; i < image->nr_segments; i++) {
152 mstart = image->segment[i].mem;
153 mend = mstart + image->segment[i].memsz;
154
155 result = kernel_ident_mapping_init(&info,
156 level4p, mstart, mend);
157
158 if (result)
159 return result;
160 }
161
162 return init_transition_pgtable(image, level4p);
163}
164
165static void set_idt(void *newidt, u16 limit)
166{
167 struct desc_ptr curidt;
168
169 /* x86-64 supports unaliged loads & stores */
170 curidt.size = limit;
171 curidt.address = (unsigned long)newidt;
172
173 __asm__ __volatile__ (
174 "lidtq %0\n"
175 : : "m" (curidt)
176 );
177};
178
179
180static void set_gdt(void *newgdt, u16 limit)
181{
182 struct desc_ptr curgdt;
183
184 /* x86-64 supports unaligned loads & stores */
185 curgdt.size = limit;
186 curgdt.address = (unsigned long)newgdt;
187
188 __asm__ __volatile__ (
189 "lgdtq %0\n"
190 : : "m" (curgdt)
191 );
192};
193
194static void load_segments(void)
195{
196 __asm__ __volatile__ (
197 "\tmovl %0,%%ds\n"
198 "\tmovl %0,%%es\n"
199 "\tmovl %0,%%ss\n"
200 "\tmovl %0,%%fs\n"
201 "\tmovl %0,%%gs\n"
202 : : "a" (__KERNEL_DS) : "memory"
203 );
204}
205
206#ifdef CONFIG_KEXEC_FILE
207/* Update purgatory as needed after various image segments have been prepared */
208static int arch_update_purgatory(struct kimage *image)
209{
210 int ret = 0;
211
212 if (!image->file_mode)
213 return 0;
214
215 /* Setup copying of backup region */
216 if (image->type == KEXEC_TYPE_CRASH) {
217 ret = kexec_purgatory_get_set_symbol(image,
218 "purgatory_backup_dest",
219 &image->arch.backup_load_addr,
220 sizeof(image->arch.backup_load_addr), 0);
221 if (ret)
222 return ret;
223
224 ret = kexec_purgatory_get_set_symbol(image,
225 "purgatory_backup_src",
226 &image->arch.backup_src_start,
227 sizeof(image->arch.backup_src_start), 0);
228 if (ret)
229 return ret;
230
231 ret = kexec_purgatory_get_set_symbol(image,
232 "purgatory_backup_sz",
233 &image->arch.backup_src_sz,
234 sizeof(image->arch.backup_src_sz), 0);
235 if (ret)
236 return ret;
237 }
238
239 return ret;
240}
241#else /* !CONFIG_KEXEC_FILE */
242static inline int arch_update_purgatory(struct kimage *image)
243{
244 return 0;
245}
246#endif /* CONFIG_KEXEC_FILE */
247
248int machine_kexec_prepare(struct kimage *image)
249{
250 unsigned long start_pgtable;
251 int result;
252
253 /* Calculate the offsets */
254 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
255
256 /* Setup the identity mapped 64bit page table */
257 result = init_pgtable(image, start_pgtable);
258 if (result)
259 return result;
260
261 /* update purgatory as needed */
262 result = arch_update_purgatory(image);
263 if (result)
264 return result;
265
266 return 0;
267}
268
269void machine_kexec_cleanup(struct kimage *image)
270{
271 free_transition_pgtable(image);
272}
273
274/*
275 * Do not allocate memory (or fail in any way) in machine_kexec().
276 * We are past the point of no return, committed to rebooting now.
277 */
278void machine_kexec(struct kimage *image)
279{
280 unsigned long page_list[PAGES_NR];
281 void *control_page;
282 int save_ftrace_enabled;
283
284#ifdef CONFIG_KEXEC_JUMP
285 if (image->preserve_context)
286 save_processor_state();
287#endif
288
289 save_ftrace_enabled = __ftrace_enabled_save();
290
291 /* Interrupts aren't acceptable while we reboot */
292 local_irq_disable();
293 hw_breakpoint_disable();
294
295 if (image->preserve_context) {
296#ifdef CONFIG_X86_IO_APIC
297 /*
298 * We need to put APICs in legacy mode so that we can
299 * get timer interrupts in second kernel. kexec/kdump
300 * paths already have calls to restore_boot_irq_mode()
301 * in one form or other. kexec jump path also need one.
302 */
303 clear_IO_APIC();
304 restore_boot_irq_mode();
305#endif
306 }
307
308 control_page = page_address(image->control_code_page) + PAGE_SIZE;
309 memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
310
311 page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
312 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
313 page_list[PA_TABLE_PAGE] =
314 (unsigned long)__pa(page_address(image->control_code_page));
315
316 if (image->type == KEXEC_TYPE_DEFAULT)
317 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
318 << PAGE_SHIFT);
319
320 /*
321 * The segment registers are funny things, they have both a
322 * visible and an invisible part. Whenever the visible part is
323 * set to a specific selector, the invisible part is loaded
324 * with from a table in memory. At no other time is the
325 * descriptor table in memory accessed.
326 *
327 * I take advantage of this here by force loading the
328 * segments, before I zap the gdt with an invalid value.
329 */
330 load_segments();
331 /*
332 * The gdt & idt are now invalid.
333 * If you want to load them you must set up your own idt & gdt.
334 */
335 set_gdt(phys_to_virt(0), 0);
336 set_idt(phys_to_virt(0), 0);
337
338 /* now call it */
339 image->start = relocate_kernel((unsigned long)image->head,
340 (unsigned long)page_list,
341 image->start,
342 image->preserve_context,
343 sme_active());
344
345#ifdef CONFIG_KEXEC_JUMP
346 if (image->preserve_context)
347 restore_processor_state();
348#endif
349
350 __ftrace_enabled_restore(save_ftrace_enabled);
351}
352
353void arch_crash_save_vmcoreinfo(void)
354{
355 VMCOREINFO_NUMBER(phys_base);
356 VMCOREINFO_SYMBOL(init_top_pgt);
357 VMCOREINFO_NUMBER(pgtable_l5_enabled);
358
359#ifdef CONFIG_NUMA
360 VMCOREINFO_SYMBOL(node_data);
361 VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
362#endif
363 vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
364 kaslr_offset());
365 VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
366}
367
368/* arch-dependent functionality related to kexec file-based syscall */
369
370#ifdef CONFIG_KEXEC_FILE
371void *arch_kexec_kernel_image_load(struct kimage *image)
372{
373 vfree(image->arch.elf_headers);
374 image->arch.elf_headers = NULL;
375
376 if (!image->fops || !image->fops->load)
377 return ERR_PTR(-ENOEXEC);
378
379 return image->fops->load(image, image->kernel_buf,
380 image->kernel_buf_len, image->initrd_buf,
381 image->initrd_buf_len, image->cmdline_buf,
382 image->cmdline_buf_len);
383}
384
385/*
386 * Apply purgatory relocations.
387 *
388 * @pi: Purgatory to be relocated.
389 * @section: Section relocations applying to.
390 * @relsec: Section containing RELAs.
391 * @symtabsec: Corresponding symtab.
392 *
393 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
394 */
395int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
396 Elf_Shdr *section, const Elf_Shdr *relsec,
397 const Elf_Shdr *symtabsec)
398{
399 unsigned int i;
400 Elf64_Rela *rel;
401 Elf64_Sym *sym;
402 void *location;
403 unsigned long address, sec_base, value;
404 const char *strtab, *name, *shstrtab;
405 const Elf_Shdr *sechdrs;
406
407 /* String & section header string table */
408 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
409 strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
410 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
411
412 rel = (void *)pi->ehdr + relsec->sh_offset;
413
414 pr_debug("Applying relocate section %s to %u\n",
415 shstrtab + relsec->sh_name, relsec->sh_info);
416
417 for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
418
419 /*
420 * rel[i].r_offset contains byte offset from beginning
421 * of section to the storage unit affected.
422 *
423 * This is location to update. This is temporary buffer
424 * where section is currently loaded. This will finally be
425 * loaded to a different address later, pointed to by
426 * ->sh_addr. kexec takes care of moving it
427 * (kexec_load_segment()).
428 */
429 location = pi->purgatory_buf;
430 location += section->sh_offset;
431 location += rel[i].r_offset;
432
433 /* Final address of the location */
434 address = section->sh_addr + rel[i].r_offset;
435
436 /*
437 * rel[i].r_info contains information about symbol table index
438 * w.r.t which relocation must be made and type of relocation
439 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
440 * these respectively.
441 */
442 sym = (void *)pi->ehdr + symtabsec->sh_offset;
443 sym += ELF64_R_SYM(rel[i].r_info);
444
445 if (sym->st_name)
446 name = strtab + sym->st_name;
447 else
448 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
449
450 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
451 name, sym->st_info, sym->st_shndx, sym->st_value,
452 sym->st_size);
453
454 if (sym->st_shndx == SHN_UNDEF) {
455 pr_err("Undefined symbol: %s\n", name);
456 return -ENOEXEC;
457 }
458
459 if (sym->st_shndx == SHN_COMMON) {
460 pr_err("symbol '%s' in common section\n", name);
461 return -ENOEXEC;
462 }
463
464 if (sym->st_shndx == SHN_ABS)
465 sec_base = 0;
466 else if (sym->st_shndx >= pi->ehdr->e_shnum) {
467 pr_err("Invalid section %d for symbol %s\n",
468 sym->st_shndx, name);
469 return -ENOEXEC;
470 } else
471 sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
472
473 value = sym->st_value;
474 value += sec_base;
475 value += rel[i].r_addend;
476
477 switch (ELF64_R_TYPE(rel[i].r_info)) {
478 case R_X86_64_NONE:
479 break;
480 case R_X86_64_64:
481 *(u64 *)location = value;
482 break;
483 case R_X86_64_32:
484 *(u32 *)location = value;
485 if (value != *(u32 *)location)
486 goto overflow;
487 break;
488 case R_X86_64_32S:
489 *(s32 *)location = value;
490 if ((s64)value != *(s32 *)location)
491 goto overflow;
492 break;
493 case R_X86_64_PC32:
494 case R_X86_64_PLT32:
495 value -= (u64)address;
496 *(u32 *)location = value;
497 break;
498 default:
499 pr_err("Unknown rela relocation: %llu\n",
500 ELF64_R_TYPE(rel[i].r_info));
501 return -ENOEXEC;
502 }
503 }
504 return 0;
505
506overflow:
507 pr_err("Overflow in relocation type %d value 0x%lx\n",
508 (int)ELF64_R_TYPE(rel[i].r_info), value);
509 return -ENOEXEC;
510}
511#endif /* CONFIG_KEXEC_FILE */
512
513static int
514kexec_mark_range(unsigned long start, unsigned long end, bool protect)
515{
516 struct page *page;
517 unsigned int nr_pages;
518
519 /*
520 * For physical range: [start, end]. We must skip the unassigned
521 * crashk resource with zero-valued "end" member.
522 */
523 if (!end || start > end)
524 return 0;
525
526 page = pfn_to_page(start >> PAGE_SHIFT);
527 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
528 if (protect)
529 return set_pages_ro(page, nr_pages);
530 else
531 return set_pages_rw(page, nr_pages);
532}
533
534static void kexec_mark_crashkres(bool protect)
535{
536 unsigned long control;
537
538 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
539
540 /* Don't touch the control code page used in crash_kexec().*/
541 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
542 /* Control code page is located in the 2nd page. */
543 kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
544 control += KEXEC_CONTROL_PAGE_SIZE;
545 kexec_mark_range(control, crashk_res.end, protect);
546}
547
548void arch_kexec_protect_crashkres(void)
549{
550 kexec_mark_crashkres(true);
551}
552
553void arch_kexec_unprotect_crashkres(void)
554{
555 kexec_mark_crashkres(false);
556}
557
558int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
559{
560 /*
561 * If SME is active we need to be sure that kexec pages are
562 * not encrypted because when we boot to the new kernel the
563 * pages won't be accessed encrypted (initially).
564 */
565 return set_memory_decrypted((unsigned long)vaddr, pages);
566}
567
568void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
569{
570 /*
571 * If SME is active we need to reset the pages back to being
572 * an encrypted mapping before freeing them.
573 */
574 set_memory_encrypted((unsigned long)vaddr, pages);
575}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * handle transition of Linux booting another kernel
4 * Copyright (C) 2002-2005 Eric Biederman <ebiederm@xmission.com>
5 */
6
7#define pr_fmt(fmt) "kexec: " fmt
8
9#include <linux/mm.h>
10#include <linux/kexec.h>
11#include <linux/string.h>
12#include <linux/gfp.h>
13#include <linux/reboot.h>
14#include <linux/numa.h>
15#include <linux/ftrace.h>
16#include <linux/io.h>
17#include <linux/suspend.h>
18#include <linux/vmalloc.h>
19#include <linux/efi.h>
20#include <linux/cc_platform.h>
21
22#include <asm/init.h>
23#include <asm/tlbflush.h>
24#include <asm/mmu_context.h>
25#include <asm/io_apic.h>
26#include <asm/debugreg.h>
27#include <asm/kexec-bzimage64.h>
28#include <asm/setup.h>
29#include <asm/set_memory.h>
30#include <asm/cpu.h>
31#include <asm/efi.h>
32
33#ifdef CONFIG_ACPI
34/*
35 * Used while adding mapping for ACPI tables.
36 * Can be reused when other iomem regions need be mapped
37 */
38struct init_pgtable_data {
39 struct x86_mapping_info *info;
40 pgd_t *level4p;
41};
42
43static int mem_region_callback(struct resource *res, void *arg)
44{
45 struct init_pgtable_data *data = arg;
46
47 return kernel_ident_mapping_init(data->info, data->level4p,
48 res->start, res->end + 1);
49}
50
51static int
52map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p)
53{
54 struct init_pgtable_data data;
55 unsigned long flags;
56 int ret;
57
58 data.info = info;
59 data.level4p = level4p;
60 flags = IORESOURCE_MEM | IORESOURCE_BUSY;
61
62 ret = walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1,
63 &data, mem_region_callback);
64 if (ret && ret != -EINVAL)
65 return ret;
66
67 /* ACPI tables could be located in ACPI Non-volatile Storage region */
68 ret = walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1,
69 &data, mem_region_callback);
70 if (ret && ret != -EINVAL)
71 return ret;
72
73 return 0;
74}
75#else
76static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; }
77#endif
78
79#ifdef CONFIG_KEXEC_FILE
80const struct kexec_file_ops * const kexec_file_loaders[] = {
81 &kexec_bzImage64_ops,
82 NULL
83};
84#endif
85
86static int
87map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p)
88{
89#ifdef CONFIG_EFI
90 unsigned long mstart, mend;
91 void *kaddr;
92 int ret;
93
94 if (!efi_enabled(EFI_BOOT))
95 return 0;
96
97 mstart = (boot_params.efi_info.efi_systab |
98 ((u64)boot_params.efi_info.efi_systab_hi<<32));
99
100 if (efi_enabled(EFI_64BIT))
101 mend = mstart + sizeof(efi_system_table_64_t);
102 else
103 mend = mstart + sizeof(efi_system_table_32_t);
104
105 if (!mstart)
106 return 0;
107
108 ret = kernel_ident_mapping_init(info, level4p, mstart, mend);
109 if (ret)
110 return ret;
111
112 kaddr = memremap(mstart, mend - mstart, MEMREMAP_WB);
113 if (!kaddr) {
114 pr_err("Could not map UEFI system table\n");
115 return -ENOMEM;
116 }
117
118 mstart = efi_config_table;
119
120 if (efi_enabled(EFI_64BIT)) {
121 efi_system_table_64_t *stbl = (efi_system_table_64_t *)kaddr;
122
123 mend = mstart + sizeof(efi_config_table_64_t) * stbl->nr_tables;
124 } else {
125 efi_system_table_32_t *stbl = (efi_system_table_32_t *)kaddr;
126
127 mend = mstart + sizeof(efi_config_table_32_t) * stbl->nr_tables;
128 }
129
130 memunmap(kaddr);
131
132 return kernel_ident_mapping_init(info, level4p, mstart, mend);
133#endif
134 return 0;
135}
136
137static void free_transition_pgtable(struct kimage *image)
138{
139 free_page((unsigned long)image->arch.p4d);
140 image->arch.p4d = NULL;
141 free_page((unsigned long)image->arch.pud);
142 image->arch.pud = NULL;
143 free_page((unsigned long)image->arch.pmd);
144 image->arch.pmd = NULL;
145 free_page((unsigned long)image->arch.pte);
146 image->arch.pte = NULL;
147}
148
149static int init_transition_pgtable(struct kimage *image, pgd_t *pgd,
150 unsigned long control_page)
151{
152 pgprot_t prot = PAGE_KERNEL_EXEC_NOENC;
153 unsigned long vaddr, paddr;
154 int result = -ENOMEM;
155 p4d_t *p4d;
156 pud_t *pud;
157 pmd_t *pmd;
158 pte_t *pte;
159
160 vaddr = (unsigned long)relocate_kernel;
161 paddr = control_page;
162 pgd += pgd_index(vaddr);
163 if (!pgd_present(*pgd)) {
164 p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
165 if (!p4d)
166 goto err;
167 image->arch.p4d = p4d;
168 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE));
169 }
170 p4d = p4d_offset(pgd, vaddr);
171 if (!p4d_present(*p4d)) {
172 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
173 if (!pud)
174 goto err;
175 image->arch.pud = pud;
176 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
177 }
178 pud = pud_offset(p4d, vaddr);
179 if (!pud_present(*pud)) {
180 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
181 if (!pmd)
182 goto err;
183 image->arch.pmd = pmd;
184 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
185 }
186 pmd = pmd_offset(pud, vaddr);
187 if (!pmd_present(*pmd)) {
188 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
189 if (!pte)
190 goto err;
191 image->arch.pte = pte;
192 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
193 }
194 pte = pte_offset_kernel(pmd, vaddr);
195
196 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
197 prot = PAGE_KERNEL_EXEC;
198
199 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot));
200 return 0;
201err:
202 return result;
203}
204
205static void *alloc_pgt_page(void *data)
206{
207 struct kimage *image = (struct kimage *)data;
208 struct page *page;
209 void *p = NULL;
210
211 page = kimage_alloc_control_pages(image, 0);
212 if (page) {
213 p = page_address(page);
214 clear_page(p);
215 }
216
217 return p;
218}
219
220static int init_pgtable(struct kimage *image, unsigned long control_page)
221{
222 struct x86_mapping_info info = {
223 .alloc_pgt_page = alloc_pgt_page,
224 .context = image,
225 .page_flag = __PAGE_KERNEL_LARGE_EXEC,
226 .kernpg_flag = _KERNPG_TABLE_NOENC,
227 };
228 unsigned long mstart, mend;
229 int result;
230 int i;
231
232 image->arch.pgd = alloc_pgt_page(image);
233 if (!image->arch.pgd)
234 return -ENOMEM;
235
236 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT)) {
237 info.page_flag |= _PAGE_ENC;
238 info.kernpg_flag |= _PAGE_ENC;
239 }
240
241 if (direct_gbpages)
242 info.direct_gbpages = true;
243
244 for (i = 0; i < nr_pfn_mapped; i++) {
245 mstart = pfn_mapped[i].start << PAGE_SHIFT;
246 mend = pfn_mapped[i].end << PAGE_SHIFT;
247
248 result = kernel_ident_mapping_init(&info, image->arch.pgd,
249 mstart, mend);
250 if (result)
251 return result;
252 }
253
254 /*
255 * segments's mem ranges could be outside 0 ~ max_pfn,
256 * for example when jump back to original kernel from kexeced kernel.
257 * or first kernel is booted with user mem map, and second kernel
258 * could be loaded out of that range.
259 */
260 for (i = 0; i < image->nr_segments; i++) {
261 mstart = image->segment[i].mem;
262 mend = mstart + image->segment[i].memsz;
263
264 result = kernel_ident_mapping_init(&info, image->arch.pgd,
265 mstart, mend);
266
267 if (result)
268 return result;
269 }
270
271 /*
272 * Prepare EFI systab and ACPI tables for kexec kernel since they are
273 * not covered by pfn_mapped.
274 */
275 result = map_efi_systab(&info, image->arch.pgd);
276 if (result)
277 return result;
278
279 result = map_acpi_tables(&info, image->arch.pgd);
280 if (result)
281 return result;
282
283 /*
284 * This must be last because the intermediate page table pages it
285 * allocates will not be control pages and may overlap the image.
286 */
287 return init_transition_pgtable(image, image->arch.pgd, control_page);
288}
289
290static void load_segments(void)
291{
292 __asm__ __volatile__ (
293 "\tmovl %0,%%ds\n"
294 "\tmovl %0,%%es\n"
295 "\tmovl %0,%%ss\n"
296 "\tmovl %0,%%fs\n"
297 "\tmovl %0,%%gs\n"
298 : : "a" (__KERNEL_DS) : "memory"
299 );
300}
301
302int machine_kexec_prepare(struct kimage *image)
303{
304 unsigned long control_page;
305 int result;
306
307 /* Calculate the offsets */
308 control_page = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
309
310 /* Setup the identity mapped 64bit page table */
311 result = init_pgtable(image, control_page);
312 if (result)
313 return result;
314
315 return 0;
316}
317
318void machine_kexec_cleanup(struct kimage *image)
319{
320 free_transition_pgtable(image);
321}
322
323/*
324 * Do not allocate memory (or fail in any way) in machine_kexec().
325 * We are past the point of no return, committed to rebooting now.
326 */
327void machine_kexec(struct kimage *image)
328{
329 unsigned long page_list[PAGES_NR];
330 unsigned int host_mem_enc_active;
331 int save_ftrace_enabled;
332 void *control_page;
333
334 /*
335 * This must be done before load_segments() since if call depth tracking
336 * is used then GS must be valid to make any function calls.
337 */
338 host_mem_enc_active = cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT);
339
340#ifdef CONFIG_KEXEC_JUMP
341 if (image->preserve_context)
342 save_processor_state();
343#endif
344
345 save_ftrace_enabled = __ftrace_enabled_save();
346
347 /* Interrupts aren't acceptable while we reboot */
348 local_irq_disable();
349 hw_breakpoint_disable();
350 cet_disable();
351
352 if (image->preserve_context) {
353#ifdef CONFIG_X86_IO_APIC
354 /*
355 * We need to put APICs in legacy mode so that we can
356 * get timer interrupts in second kernel. kexec/kdump
357 * paths already have calls to restore_boot_irq_mode()
358 * in one form or other. kexec jump path also need one.
359 */
360 clear_IO_APIC();
361 restore_boot_irq_mode();
362#endif
363 }
364
365 control_page = page_address(image->control_code_page);
366 __memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
367
368 page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
369 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
370 page_list[PA_TABLE_PAGE] = (unsigned long)__pa(image->arch.pgd);
371
372 if (image->type == KEXEC_TYPE_DEFAULT)
373 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
374 << PAGE_SHIFT);
375
376 /*
377 * The segment registers are funny things, they have both a
378 * visible and an invisible part. Whenever the visible part is
379 * set to a specific selector, the invisible part is loaded
380 * with from a table in memory. At no other time is the
381 * descriptor table in memory accessed.
382 *
383 * I take advantage of this here by force loading the
384 * segments, before I zap the gdt with an invalid value.
385 */
386 load_segments();
387 /*
388 * The gdt & idt are now invalid.
389 * If you want to load them you must set up your own idt & gdt.
390 */
391 native_idt_invalidate();
392 native_gdt_invalidate();
393
394 /* now call it */
395 image->start = relocate_kernel((unsigned long)image->head,
396 (unsigned long)page_list,
397 image->start,
398 image->preserve_context,
399 host_mem_enc_active);
400
401#ifdef CONFIG_KEXEC_JUMP
402 if (image->preserve_context)
403 restore_processor_state();
404#endif
405
406 __ftrace_enabled_restore(save_ftrace_enabled);
407}
408
409/* arch-dependent functionality related to kexec file-based syscall */
410
411#ifdef CONFIG_KEXEC_FILE
412/*
413 * Apply purgatory relocations.
414 *
415 * @pi: Purgatory to be relocated.
416 * @section: Section relocations applying to.
417 * @relsec: Section containing RELAs.
418 * @symtabsec: Corresponding symtab.
419 *
420 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
421 */
422int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
423 Elf_Shdr *section, const Elf_Shdr *relsec,
424 const Elf_Shdr *symtabsec)
425{
426 unsigned int i;
427 Elf64_Rela *rel;
428 Elf64_Sym *sym;
429 void *location;
430 unsigned long address, sec_base, value;
431 const char *strtab, *name, *shstrtab;
432 const Elf_Shdr *sechdrs;
433
434 /* String & section header string table */
435 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
436 strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset;
437 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
438
439 rel = (void *)pi->ehdr + relsec->sh_offset;
440
441 pr_debug("Applying relocate section %s to %u\n",
442 shstrtab + relsec->sh_name, relsec->sh_info);
443
444 for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) {
445
446 /*
447 * rel[i].r_offset contains byte offset from beginning
448 * of section to the storage unit affected.
449 *
450 * This is location to update. This is temporary buffer
451 * where section is currently loaded. This will finally be
452 * loaded to a different address later, pointed to by
453 * ->sh_addr. kexec takes care of moving it
454 * (kexec_load_segment()).
455 */
456 location = pi->purgatory_buf;
457 location += section->sh_offset;
458 location += rel[i].r_offset;
459
460 /* Final address of the location */
461 address = section->sh_addr + rel[i].r_offset;
462
463 /*
464 * rel[i].r_info contains information about symbol table index
465 * w.r.t which relocation must be made and type of relocation
466 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
467 * these respectively.
468 */
469 sym = (void *)pi->ehdr + symtabsec->sh_offset;
470 sym += ELF64_R_SYM(rel[i].r_info);
471
472 if (sym->st_name)
473 name = strtab + sym->st_name;
474 else
475 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
476
477 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
478 name, sym->st_info, sym->st_shndx, sym->st_value,
479 sym->st_size);
480
481 if (sym->st_shndx == SHN_UNDEF) {
482 pr_err("Undefined symbol: %s\n", name);
483 return -ENOEXEC;
484 }
485
486 if (sym->st_shndx == SHN_COMMON) {
487 pr_err("symbol '%s' in common section\n", name);
488 return -ENOEXEC;
489 }
490
491 if (sym->st_shndx == SHN_ABS)
492 sec_base = 0;
493 else if (sym->st_shndx >= pi->ehdr->e_shnum) {
494 pr_err("Invalid section %d for symbol %s\n",
495 sym->st_shndx, name);
496 return -ENOEXEC;
497 } else
498 sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
499
500 value = sym->st_value;
501 value += sec_base;
502 value += rel[i].r_addend;
503
504 switch (ELF64_R_TYPE(rel[i].r_info)) {
505 case R_X86_64_NONE:
506 break;
507 case R_X86_64_64:
508 *(u64 *)location = value;
509 break;
510 case R_X86_64_32:
511 *(u32 *)location = value;
512 if (value != *(u32 *)location)
513 goto overflow;
514 break;
515 case R_X86_64_32S:
516 *(s32 *)location = value;
517 if ((s64)value != *(s32 *)location)
518 goto overflow;
519 break;
520 case R_X86_64_PC32:
521 case R_X86_64_PLT32:
522 value -= (u64)address;
523 *(u32 *)location = value;
524 break;
525 default:
526 pr_err("Unknown rela relocation: %llu\n",
527 ELF64_R_TYPE(rel[i].r_info));
528 return -ENOEXEC;
529 }
530 }
531 return 0;
532
533overflow:
534 pr_err("Overflow in relocation type %d value 0x%lx\n",
535 (int)ELF64_R_TYPE(rel[i].r_info), value);
536 return -ENOEXEC;
537}
538
539int arch_kimage_file_post_load_cleanup(struct kimage *image)
540{
541 vfree(image->elf_headers);
542 image->elf_headers = NULL;
543 image->elf_headers_sz = 0;
544
545 return kexec_image_post_load_cleanup_default(image);
546}
547#endif /* CONFIG_KEXEC_FILE */
548
549#ifdef CONFIG_CRASH_DUMP
550
551static int
552kexec_mark_range(unsigned long start, unsigned long end, bool protect)
553{
554 struct page *page;
555 unsigned int nr_pages;
556
557 /*
558 * For physical range: [start, end]. We must skip the unassigned
559 * crashk resource with zero-valued "end" member.
560 */
561 if (!end || start > end)
562 return 0;
563
564 page = pfn_to_page(start >> PAGE_SHIFT);
565 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
566 if (protect)
567 return set_pages_ro(page, nr_pages);
568 else
569 return set_pages_rw(page, nr_pages);
570}
571
572static void kexec_mark_crashkres(bool protect)
573{
574 unsigned long control;
575
576 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
577
578 /* Don't touch the control code page used in crash_kexec().*/
579 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
580 kexec_mark_range(crashk_res.start, control - 1, protect);
581 control += KEXEC_CONTROL_PAGE_SIZE;
582 kexec_mark_range(control, crashk_res.end, protect);
583}
584
585void arch_kexec_protect_crashkres(void)
586{
587 kexec_mark_crashkres(true);
588}
589
590void arch_kexec_unprotect_crashkres(void)
591{
592 kexec_mark_crashkres(false);
593}
594#endif
595
596/*
597 * During a traditional boot under SME, SME will encrypt the kernel,
598 * so the SME kexec kernel also needs to be un-encrypted in order to
599 * replicate a normal SME boot.
600 *
601 * During a traditional boot under SEV, the kernel has already been
602 * loaded encrypted, so the SEV kexec kernel needs to be encrypted in
603 * order to replicate a normal SEV boot.
604 */
605int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp)
606{
607 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
608 return 0;
609
610 /*
611 * If host memory encryption is active we need to be sure that kexec
612 * pages are not encrypted because when we boot to the new kernel the
613 * pages won't be accessed encrypted (initially).
614 */
615 return set_memory_decrypted((unsigned long)vaddr, pages);
616}
617
618void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages)
619{
620 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT))
621 return;
622
623 /*
624 * If host memory encryption is active we need to reset the pages back
625 * to being an encrypted mapping before freeing them.
626 */
627 set_memory_encrypted((unsigned long)vaddr, pages);
628}