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/*
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
31#ifdef CONFIG_KEXEC_FILE
32static struct kexec_file_ops *kexec_file_loaders[] = {
33 &kexec_bzImage64_ops,
34};
35#endif
36
37static void free_transition_pgtable(struct kimage *image)
38{
39 free_page((unsigned long)image->arch.pud);
40 free_page((unsigned long)image->arch.pmd);
41 free_page((unsigned long)image->arch.pte);
42}
43
44static int init_transition_pgtable(struct kimage *image, pgd_t *pgd)
45{
46 pud_t *pud;
47 pmd_t *pmd;
48 pte_t *pte;
49 unsigned long vaddr, paddr;
50 int result = -ENOMEM;
51
52 vaddr = (unsigned long)relocate_kernel;
53 paddr = __pa(page_address(image->control_code_page)+PAGE_SIZE);
54 pgd += pgd_index(vaddr);
55 if (!pgd_present(*pgd)) {
56 pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
57 if (!pud)
58 goto err;
59 image->arch.pud = pud;
60 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
61 }
62 pud = pud_offset(pgd, vaddr);
63 if (!pud_present(*pud)) {
64 pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
65 if (!pmd)
66 goto err;
67 image->arch.pmd = pmd;
68 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
69 }
70 pmd = pmd_offset(pud, vaddr);
71 if (!pmd_present(*pmd)) {
72 pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
73 if (!pte)
74 goto err;
75 image->arch.pte = pte;
76 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
77 }
78 pte = pte_offset_kernel(pmd, vaddr);
79 set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL_EXEC));
80 return 0;
81err:
82 free_transition_pgtable(image);
83 return result;
84}
85
86static void *alloc_pgt_page(void *data)
87{
88 struct kimage *image = (struct kimage *)data;
89 struct page *page;
90 void *p = NULL;
91
92 page = kimage_alloc_control_pages(image, 0);
93 if (page) {
94 p = page_address(page);
95 clear_page(p);
96 }
97
98 return p;
99}
100
101static int init_pgtable(struct kimage *image, unsigned long start_pgtable)
102{
103 struct x86_mapping_info info = {
104 .alloc_pgt_page = alloc_pgt_page,
105 .context = image,
106 .pmd_flag = __PAGE_KERNEL_LARGE_EXEC,
107 };
108 unsigned long mstart, mend;
109 pgd_t *level4p;
110 int result;
111 int i;
112
113 level4p = (pgd_t *)__va(start_pgtable);
114 clear_page(level4p);
115 for (i = 0; i < nr_pfn_mapped; i++) {
116 mstart = pfn_mapped[i].start << PAGE_SHIFT;
117 mend = pfn_mapped[i].end << PAGE_SHIFT;
118
119 result = kernel_ident_mapping_init(&info,
120 level4p, mstart, mend);
121 if (result)
122 return result;
123 }
124
125 /*
126 * segments's mem ranges could be outside 0 ~ max_pfn,
127 * for example when jump back to original kernel from kexeced kernel.
128 * or first kernel is booted with user mem map, and second kernel
129 * could be loaded out of that range.
130 */
131 for (i = 0; i < image->nr_segments; i++) {
132 mstart = image->segment[i].mem;
133 mend = mstart + image->segment[i].memsz;
134
135 result = kernel_ident_mapping_init(&info,
136 level4p, mstart, mend);
137
138 if (result)
139 return result;
140 }
141
142 return init_transition_pgtable(image, level4p);
143}
144
145static void set_idt(void *newidt, u16 limit)
146{
147 struct desc_ptr curidt;
148
149 /* x86-64 supports unaliged loads & stores */
150 curidt.size = limit;
151 curidt.address = (unsigned long)newidt;
152
153 __asm__ __volatile__ (
154 "lidtq %0\n"
155 : : "m" (curidt)
156 );
157};
158
159
160static void set_gdt(void *newgdt, u16 limit)
161{
162 struct desc_ptr curgdt;
163
164 /* x86-64 supports unaligned loads & stores */
165 curgdt.size = limit;
166 curgdt.address = (unsigned long)newgdt;
167
168 __asm__ __volatile__ (
169 "lgdtq %0\n"
170 : : "m" (curgdt)
171 );
172};
173
174static void load_segments(void)
175{
176 __asm__ __volatile__ (
177 "\tmovl %0,%%ds\n"
178 "\tmovl %0,%%es\n"
179 "\tmovl %0,%%ss\n"
180 "\tmovl %0,%%fs\n"
181 "\tmovl %0,%%gs\n"
182 : : "a" (__KERNEL_DS) : "memory"
183 );
184}
185
186#ifdef CONFIG_KEXEC_FILE
187/* Update purgatory as needed after various image segments have been prepared */
188static int arch_update_purgatory(struct kimage *image)
189{
190 int ret = 0;
191
192 if (!image->file_mode)
193 return 0;
194
195 /* Setup copying of backup region */
196 if (image->type == KEXEC_TYPE_CRASH) {
197 ret = kexec_purgatory_get_set_symbol(image, "backup_dest",
198 &image->arch.backup_load_addr,
199 sizeof(image->arch.backup_load_addr), 0);
200 if (ret)
201 return ret;
202
203 ret = kexec_purgatory_get_set_symbol(image, "backup_src",
204 &image->arch.backup_src_start,
205 sizeof(image->arch.backup_src_start), 0);
206 if (ret)
207 return ret;
208
209 ret = kexec_purgatory_get_set_symbol(image, "backup_sz",
210 &image->arch.backup_src_sz,
211 sizeof(image->arch.backup_src_sz), 0);
212 if (ret)
213 return ret;
214 }
215
216 return ret;
217}
218#else /* !CONFIG_KEXEC_FILE */
219static inline int arch_update_purgatory(struct kimage *image)
220{
221 return 0;
222}
223#endif /* CONFIG_KEXEC_FILE */
224
225int machine_kexec_prepare(struct kimage *image)
226{
227 unsigned long start_pgtable;
228 int result;
229
230 /* Calculate the offsets */
231 start_pgtable = page_to_pfn(image->control_code_page) << PAGE_SHIFT;
232
233 /* Setup the identity mapped 64bit page table */
234 result = init_pgtable(image, start_pgtable);
235 if (result)
236 return result;
237
238 /* update purgatory as needed */
239 result = arch_update_purgatory(image);
240 if (result)
241 return result;
242
243 return 0;
244}
245
246void machine_kexec_cleanup(struct kimage *image)
247{
248 free_transition_pgtable(image);
249}
250
251/*
252 * Do not allocate memory (or fail in any way) in machine_kexec().
253 * We are past the point of no return, committed to rebooting now.
254 */
255void machine_kexec(struct kimage *image)
256{
257 unsigned long page_list[PAGES_NR];
258 void *control_page;
259 int save_ftrace_enabled;
260
261#ifdef CONFIG_KEXEC_JUMP
262 if (image->preserve_context)
263 save_processor_state();
264#endif
265
266 save_ftrace_enabled = __ftrace_enabled_save();
267
268 /* Interrupts aren't acceptable while we reboot */
269 local_irq_disable();
270 hw_breakpoint_disable();
271
272 if (image->preserve_context) {
273#ifdef CONFIG_X86_IO_APIC
274 /*
275 * We need to put APICs in legacy mode so that we can
276 * get timer interrupts in second kernel. kexec/kdump
277 * paths already have calls to disable_IO_APIC() in
278 * one form or other. kexec jump path also need
279 * one.
280 */
281 disable_IO_APIC();
282#endif
283 }
284
285 control_page = page_address(image->control_code_page) + PAGE_SIZE;
286 memcpy(control_page, relocate_kernel, KEXEC_CONTROL_CODE_MAX_SIZE);
287
288 page_list[PA_CONTROL_PAGE] = virt_to_phys(control_page);
289 page_list[VA_CONTROL_PAGE] = (unsigned long)control_page;
290 page_list[PA_TABLE_PAGE] =
291 (unsigned long)__pa(page_address(image->control_code_page));
292
293 if (image->type == KEXEC_TYPE_DEFAULT)
294 page_list[PA_SWAP_PAGE] = (page_to_pfn(image->swap_page)
295 << PAGE_SHIFT);
296
297 /*
298 * The segment registers are funny things, they have both a
299 * visible and an invisible part. Whenever the visible part is
300 * set to a specific selector, the invisible part is loaded
301 * with from a table in memory. At no other time is the
302 * descriptor table in memory accessed.
303 *
304 * I take advantage of this here by force loading the
305 * segments, before I zap the gdt with an invalid value.
306 */
307 load_segments();
308 /*
309 * The gdt & idt are now invalid.
310 * If you want to load them you must set up your own idt & gdt.
311 */
312 set_gdt(phys_to_virt(0), 0);
313 set_idt(phys_to_virt(0), 0);
314
315 /* now call it */
316 image->start = relocate_kernel((unsigned long)image->head,
317 (unsigned long)page_list,
318 image->start,
319 image->preserve_context);
320
321#ifdef CONFIG_KEXEC_JUMP
322 if (image->preserve_context)
323 restore_processor_state();
324#endif
325
326 __ftrace_enabled_restore(save_ftrace_enabled);
327}
328
329void arch_crash_save_vmcoreinfo(void)
330{
331 VMCOREINFO_NUMBER(phys_base);
332 VMCOREINFO_SYMBOL(init_level4_pgt);
333
334#ifdef CONFIG_NUMA
335 VMCOREINFO_SYMBOL(node_data);
336 VMCOREINFO_LENGTH(node_data, MAX_NUMNODES);
337#endif
338 vmcoreinfo_append_str("KERNELOFFSET=%lx\n",
339 kaslr_offset());
340 VMCOREINFO_NUMBER(KERNEL_IMAGE_SIZE);
341}
342
343/* arch-dependent functionality related to kexec file-based syscall */
344
345#ifdef CONFIG_KEXEC_FILE
346int arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
347 unsigned long buf_len)
348{
349 int i, ret = -ENOEXEC;
350 struct kexec_file_ops *fops;
351
352 for (i = 0; i < ARRAY_SIZE(kexec_file_loaders); i++) {
353 fops = kexec_file_loaders[i];
354 if (!fops || !fops->probe)
355 continue;
356
357 ret = fops->probe(buf, buf_len);
358 if (!ret) {
359 image->fops = fops;
360 return ret;
361 }
362 }
363
364 return ret;
365}
366
367void *arch_kexec_kernel_image_load(struct kimage *image)
368{
369 vfree(image->arch.elf_headers);
370 image->arch.elf_headers = NULL;
371
372 if (!image->fops || !image->fops->load)
373 return ERR_PTR(-ENOEXEC);
374
375 return image->fops->load(image, image->kernel_buf,
376 image->kernel_buf_len, image->initrd_buf,
377 image->initrd_buf_len, image->cmdline_buf,
378 image->cmdline_buf_len);
379}
380
381int arch_kimage_file_post_load_cleanup(struct kimage *image)
382{
383 if (!image->fops || !image->fops->cleanup)
384 return 0;
385
386 return image->fops->cleanup(image->image_loader_data);
387}
388
389#ifdef CONFIG_KEXEC_VERIFY_SIG
390int arch_kexec_kernel_verify_sig(struct kimage *image, void *kernel,
391 unsigned long kernel_len)
392{
393 if (!image->fops || !image->fops->verify_sig) {
394 pr_debug("kernel loader does not support signature verification.");
395 return -EKEYREJECTED;
396 }
397
398 return image->fops->verify_sig(kernel, kernel_len);
399}
400#endif
401
402/*
403 * Apply purgatory relocations.
404 *
405 * ehdr: Pointer to elf headers
406 * sechdrs: Pointer to section headers.
407 * relsec: section index of SHT_RELA section.
408 *
409 * TODO: Some of the code belongs to generic code. Move that in kexec.c.
410 */
411int arch_kexec_apply_relocations_add(const Elf64_Ehdr *ehdr,
412 Elf64_Shdr *sechdrs, unsigned int relsec)
413{
414 unsigned int i;
415 Elf64_Rela *rel;
416 Elf64_Sym *sym;
417 void *location;
418 Elf64_Shdr *section, *symtabsec;
419 unsigned long address, sec_base, value;
420 const char *strtab, *name, *shstrtab;
421
422 /*
423 * ->sh_offset has been modified to keep the pointer to section
424 * contents in memory
425 */
426 rel = (void *)sechdrs[relsec].sh_offset;
427
428 /* Section to which relocations apply */
429 section = &sechdrs[sechdrs[relsec].sh_info];
430
431 pr_debug("Applying relocate section %u to %u\n", relsec,
432 sechdrs[relsec].sh_info);
433
434 /* Associated symbol table */
435 symtabsec = &sechdrs[sechdrs[relsec].sh_link];
436
437 /* String table */
438 if (symtabsec->sh_link >= ehdr->e_shnum) {
439 /* Invalid strtab section number */
440 pr_err("Invalid string table section index %d\n",
441 symtabsec->sh_link);
442 return -ENOEXEC;
443 }
444
445 strtab = (char *)sechdrs[symtabsec->sh_link].sh_offset;
446
447 /* section header string table */
448 shstrtab = (char *)sechdrs[ehdr->e_shstrndx].sh_offset;
449
450 for (i = 0; i < sechdrs[relsec].sh_size / sizeof(*rel); i++) {
451
452 /*
453 * rel[i].r_offset contains byte offset from beginning
454 * of section to the storage unit affected.
455 *
456 * This is location to update (->sh_offset). This is temporary
457 * buffer where section is currently loaded. This will finally
458 * be loaded to a different address later, pointed to by
459 * ->sh_addr. kexec takes care of moving it
460 * (kexec_load_segment()).
461 */
462 location = (void *)(section->sh_offset + rel[i].r_offset);
463
464 /* Final address of the location */
465 address = section->sh_addr + rel[i].r_offset;
466
467 /*
468 * rel[i].r_info contains information about symbol table index
469 * w.r.t which relocation must be made and type of relocation
470 * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get
471 * these respectively.
472 */
473 sym = (Elf64_Sym *)symtabsec->sh_offset +
474 ELF64_R_SYM(rel[i].r_info);
475
476 if (sym->st_name)
477 name = strtab + sym->st_name;
478 else
479 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
480
481 pr_debug("Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n",
482 name, sym->st_info, sym->st_shndx, sym->st_value,
483 sym->st_size);
484
485 if (sym->st_shndx == SHN_UNDEF) {
486 pr_err("Undefined symbol: %s\n", name);
487 return -ENOEXEC;
488 }
489
490 if (sym->st_shndx == SHN_COMMON) {
491 pr_err("symbol '%s' in common section\n", name);
492 return -ENOEXEC;
493 }
494
495 if (sym->st_shndx == SHN_ABS)
496 sec_base = 0;
497 else if (sym->st_shndx >= ehdr->e_shnum) {
498 pr_err("Invalid section %d for symbol %s\n",
499 sym->st_shndx, name);
500 return -ENOEXEC;
501 } else
502 sec_base = sechdrs[sym->st_shndx].sh_addr;
503
504 value = sym->st_value;
505 value += sec_base;
506 value += rel[i].r_addend;
507
508 switch (ELF64_R_TYPE(rel[i].r_info)) {
509 case R_X86_64_NONE:
510 break;
511 case R_X86_64_64:
512 *(u64 *)location = value;
513 break;
514 case R_X86_64_32:
515 *(u32 *)location = value;
516 if (value != *(u32 *)location)
517 goto overflow;
518 break;
519 case R_X86_64_32S:
520 *(s32 *)location = value;
521 if ((s64)value != *(s32 *)location)
522 goto overflow;
523 break;
524 case R_X86_64_PC32:
525 value -= (u64)address;
526 *(u32 *)location = value;
527 break;
528 default:
529 pr_err("Unknown rela relocation: %llu\n",
530 ELF64_R_TYPE(rel[i].r_info));
531 return -ENOEXEC;
532 }
533 }
534 return 0;
535
536overflow:
537 pr_err("Overflow in relocation type %d value 0x%lx\n",
538 (int)ELF64_R_TYPE(rel[i].r_info), value);
539 return -ENOEXEC;
540}
541#endif /* CONFIG_KEXEC_FILE */
542
543static int
544kexec_mark_range(unsigned long start, unsigned long end, bool protect)
545{
546 struct page *page;
547 unsigned int nr_pages;
548
549 /*
550 * For physical range: [start, end]. We must skip the unassigned
551 * crashk resource with zero-valued "end" member.
552 */
553 if (!end || start > end)
554 return 0;
555
556 page = pfn_to_page(start >> PAGE_SHIFT);
557 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
558 if (protect)
559 return set_pages_ro(page, nr_pages);
560 else
561 return set_pages_rw(page, nr_pages);
562}
563
564static void kexec_mark_crashkres(bool protect)
565{
566 unsigned long control;
567
568 kexec_mark_range(crashk_low_res.start, crashk_low_res.end, protect);
569
570 /* Don't touch the control code page used in crash_kexec().*/
571 control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page));
572 /* Control code page is located in the 2nd page. */
573 kexec_mark_range(crashk_res.start, control + PAGE_SIZE - 1, protect);
574 control += KEXEC_CONTROL_PAGE_SIZE;
575 kexec_mark_range(control, crashk_res.end, protect);
576}
577
578void arch_kexec_protect_crashkres(void)
579{
580 kexec_mark_crashkres(true);
581}
582
583void arch_kexec_unprotect_crashkres(void)
584{
585 kexec_mark_crashkres(false);
586}