| 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 | #include <asm/processor.h> | 
|---|
| 33 |  | 
|---|
| 34 | #ifdef CONFIG_ACPI | 
|---|
| 35 | /* | 
|---|
| 36 | * Used while adding mapping for ACPI tables. | 
|---|
| 37 | * Can be reused when other iomem regions need be mapped | 
|---|
| 38 | */ | 
|---|
| 39 | struct init_pgtable_data { | 
|---|
| 40 | struct x86_mapping_info *info; | 
|---|
| 41 | pgd_t *level4p; | 
|---|
| 42 | }; | 
|---|
| 43 |  | 
|---|
| 44 | static int mem_region_callback(struct resource *res, void *arg) | 
|---|
| 45 | { | 
|---|
| 46 | struct init_pgtable_data *data = arg; | 
|---|
| 47 |  | 
|---|
| 48 | return kernel_ident_mapping_init(info: data->info, pgd_page: data->level4p, | 
|---|
| 49 | pstart: res->start, pend: res->end + 1); | 
|---|
| 50 | } | 
|---|
| 51 |  | 
|---|
| 52 | static int | 
|---|
| 53 | map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) | 
|---|
| 54 | { | 
|---|
| 55 | struct init_pgtable_data data; | 
|---|
| 56 | unsigned long flags; | 
|---|
| 57 | int ret; | 
|---|
| 58 |  | 
|---|
| 59 | data.info = info; | 
|---|
| 60 | data.level4p = level4p; | 
|---|
| 61 | flags = IORESOURCE_MEM | IORESOURCE_BUSY; | 
|---|
| 62 |  | 
|---|
| 63 | ret = walk_iomem_res_desc(desc: IORES_DESC_ACPI_TABLES, flags, start: 0, end: -1, | 
|---|
| 64 | arg: &data, func: mem_region_callback); | 
|---|
| 65 | if (ret && ret != -EINVAL) | 
|---|
| 66 | return ret; | 
|---|
| 67 |  | 
|---|
| 68 | /* ACPI tables could be located in ACPI Non-volatile Storage region */ | 
|---|
| 69 | ret = walk_iomem_res_desc(desc: IORES_DESC_ACPI_NV_STORAGE, flags, start: 0, end: -1, | 
|---|
| 70 | arg: &data, func: mem_region_callback); | 
|---|
| 71 | if (ret && ret != -EINVAL) | 
|---|
| 72 | return ret; | 
|---|
| 73 |  | 
|---|
| 74 | return 0; | 
|---|
| 75 | } | 
|---|
| 76 | #else | 
|---|
| 77 | static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; } | 
|---|
| 78 | #endif | 
|---|
| 79 |  | 
|---|
| 80 | static int map_mmio_serial(struct x86_mapping_info *info, pgd_t *level4p) | 
|---|
| 81 | { | 
|---|
| 82 | unsigned long mstart, mend; | 
|---|
| 83 |  | 
|---|
| 84 | if (!kexec_debug_8250_mmio32) | 
|---|
| 85 | return 0; | 
|---|
| 86 |  | 
|---|
| 87 | mstart = kexec_debug_8250_mmio32 & PAGE_MASK; | 
|---|
| 88 | mend = (kexec_debug_8250_mmio32 + PAGE_SIZE + 23) & PAGE_MASK; | 
|---|
| 89 | pr_info( "Map PCI serial at %lx - %lx\n", mstart, mend); | 
|---|
| 90 | return kernel_ident_mapping_init(info, pgd_page: level4p, pstart: mstart, pend: mend); | 
|---|
| 91 | } | 
|---|
| 92 |  | 
|---|
| 93 | #ifdef CONFIG_KEXEC_FILE | 
|---|
| 94 | const struct kexec_file_ops * const kexec_file_loaders[] = { | 
|---|
| 95 | &kexec_bzImage64_ops, | 
|---|
| 96 | NULL | 
|---|
| 97 | }; | 
|---|
| 98 | #endif | 
|---|
| 99 |  | 
|---|
| 100 | static int | 
|---|
| 101 | map_efi_systab(struct x86_mapping_info *info, pgd_t *level4p) | 
|---|
| 102 | { | 
|---|
| 103 | #ifdef CONFIG_EFI | 
|---|
| 104 | unsigned long mstart, mend; | 
|---|
| 105 | void *kaddr; | 
|---|
| 106 | int ret; | 
|---|
| 107 |  | 
|---|
| 108 | if (!efi_enabled(EFI_BOOT)) | 
|---|
| 109 | return 0; | 
|---|
| 110 |  | 
|---|
| 111 | mstart = (boot_params.efi_info.efi_systab | | 
|---|
| 112 | ((u64)boot_params.efi_info.efi_systab_hi<<32)); | 
|---|
| 113 |  | 
|---|
| 114 | if (efi_enabled(EFI_64BIT)) | 
|---|
| 115 | mend = mstart + sizeof(efi_system_table_64_t); | 
|---|
| 116 | else | 
|---|
| 117 | mend = mstart + sizeof(efi_system_table_32_t); | 
|---|
| 118 |  | 
|---|
| 119 | if (!mstart) | 
|---|
| 120 | return 0; | 
|---|
| 121 |  | 
|---|
| 122 | ret = kernel_ident_mapping_init(info, pgd_page: level4p, pstart: mstart, pend: mend); | 
|---|
| 123 | if (ret) | 
|---|
| 124 | return ret; | 
|---|
| 125 |  | 
|---|
| 126 | kaddr = memremap(offset: mstart, size: mend - mstart, flags: MEMREMAP_WB); | 
|---|
| 127 | if (!kaddr) { | 
|---|
| 128 | pr_err( "Could not map UEFI system table\n"); | 
|---|
| 129 | return -ENOMEM; | 
|---|
| 130 | } | 
|---|
| 131 |  | 
|---|
| 132 | mstart = efi_config_table; | 
|---|
| 133 |  | 
|---|
| 134 | if (efi_enabled(EFI_64BIT)) { | 
|---|
| 135 | efi_system_table_64_t *stbl = (efi_system_table_64_t *)kaddr; | 
|---|
| 136 |  | 
|---|
| 137 | mend = mstart + sizeof(efi_config_table_64_t) * stbl->nr_tables; | 
|---|
| 138 | } else { | 
|---|
| 139 | efi_system_table_32_t *stbl = (efi_system_table_32_t *)kaddr; | 
|---|
| 140 |  | 
|---|
| 141 | mend = mstart + sizeof(efi_config_table_32_t) * stbl->nr_tables; | 
|---|
| 142 | } | 
|---|
| 143 |  | 
|---|
| 144 | memunmap(addr: kaddr); | 
|---|
| 145 |  | 
|---|
| 146 | return kernel_ident_mapping_init(info, pgd_page: level4p, pstart: mstart, pend: mend); | 
|---|
| 147 | #endif | 
|---|
| 148 | return 0; | 
|---|
| 149 | } | 
|---|
| 150 |  | 
|---|
| 151 | static void free_transition_pgtable(struct kimage *image) | 
|---|
| 152 | { | 
|---|
| 153 | free_page((unsigned long)image->arch.p4d); | 
|---|
| 154 | image->arch.p4d = NULL; | 
|---|
| 155 | free_page((unsigned long)image->arch.pud); | 
|---|
| 156 | image->arch.pud = NULL; | 
|---|
| 157 | free_page((unsigned long)image->arch.pmd); | 
|---|
| 158 | image->arch.pmd = NULL; | 
|---|
| 159 | free_page((unsigned long)image->arch.pte); | 
|---|
| 160 | image->arch.pte = NULL; | 
|---|
| 161 | } | 
|---|
| 162 |  | 
|---|
| 163 | static int init_transition_pgtable(struct kimage *image, pgd_t *pgd, | 
|---|
| 164 | unsigned long control_page) | 
|---|
| 165 | { | 
|---|
| 166 | pgprot_t prot = PAGE_KERNEL_EXEC_NOENC; | 
|---|
| 167 | unsigned long vaddr, paddr; | 
|---|
| 168 | int result = -ENOMEM; | 
|---|
| 169 | p4d_t *p4d; | 
|---|
| 170 | pud_t *pud; | 
|---|
| 171 | pmd_t *pmd; | 
|---|
| 172 | pte_t *pte; | 
|---|
| 173 |  | 
|---|
| 174 | /* | 
|---|
| 175 | * For the transition to the identity mapped page tables, the control | 
|---|
| 176 | * code page also needs to be mapped at the virtual address it starts | 
|---|
| 177 | * off running from. | 
|---|
| 178 | */ | 
|---|
| 179 | vaddr = (unsigned long)__va(control_page); | 
|---|
| 180 | paddr = control_page; | 
|---|
| 181 | pgd += pgd_index(vaddr); | 
|---|
| 182 | if (!pgd_present(pgd: *pgd)) { | 
|---|
| 183 | p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL); | 
|---|
| 184 | if (!p4d) | 
|---|
| 185 | goto err; | 
|---|
| 186 | image->arch.p4d = p4d; | 
|---|
| 187 | set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE)); | 
|---|
| 188 | } | 
|---|
| 189 | p4d = p4d_offset(pgd, address: vaddr); | 
|---|
| 190 | if (!p4d_present(p4d: *p4d)) { | 
|---|
| 191 | pud = (pud_t *)get_zeroed_page(GFP_KERNEL); | 
|---|
| 192 | if (!pud) | 
|---|
| 193 | goto err; | 
|---|
| 194 | image->arch.pud = pud; | 
|---|
| 195 | set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE)); | 
|---|
| 196 | } | 
|---|
| 197 | pud = pud_offset(p4d, address: vaddr); | 
|---|
| 198 | if (!pud_present(pud: *pud)) { | 
|---|
| 199 | pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL); | 
|---|
| 200 | if (!pmd) | 
|---|
| 201 | goto err; | 
|---|
| 202 | image->arch.pmd = pmd; | 
|---|
| 203 | set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE)); | 
|---|
| 204 | } | 
|---|
| 205 | pmd = pmd_offset(pud, address: vaddr); | 
|---|
| 206 | if (!pmd_present(pmd: *pmd)) { | 
|---|
| 207 | pte = (pte_t *)get_zeroed_page(GFP_KERNEL); | 
|---|
| 208 | if (!pte) | 
|---|
| 209 | goto err; | 
|---|
| 210 | image->arch.pte = pte; | 
|---|
| 211 | set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE)); | 
|---|
| 212 | } | 
|---|
| 213 | pte = pte_offset_kernel(pmd, address: vaddr); | 
|---|
| 214 |  | 
|---|
| 215 | if (cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT)) | 
|---|
| 216 | prot = PAGE_KERNEL_EXEC; | 
|---|
| 217 |  | 
|---|
| 218 | set_pte(pte, pfn_pte(paddr >> PAGE_SHIFT, prot)); | 
|---|
| 219 | return 0; | 
|---|
| 220 | err: | 
|---|
| 221 | return result; | 
|---|
| 222 | } | 
|---|
| 223 |  | 
|---|
| 224 | static void *alloc_pgt_page(void *data) | 
|---|
| 225 | { | 
|---|
| 226 | struct kimage *image = (struct kimage *)data; | 
|---|
| 227 | struct page *page; | 
|---|
| 228 | void *p = NULL; | 
|---|
| 229 |  | 
|---|
| 230 | page = kimage_alloc_control_pages(image, order: 0); | 
|---|
| 231 | if (page) { | 
|---|
| 232 | p = page_address(page); | 
|---|
| 233 | clear_page(page: p); | 
|---|
| 234 | } | 
|---|
| 235 |  | 
|---|
| 236 | return p; | 
|---|
| 237 | } | 
|---|
| 238 |  | 
|---|
| 239 | static int init_pgtable(struct kimage *image, unsigned long control_page) | 
|---|
| 240 | { | 
|---|
| 241 | struct x86_mapping_info info = { | 
|---|
| 242 | .alloc_pgt_page	= alloc_pgt_page, | 
|---|
| 243 | .context	= image, | 
|---|
| 244 | .page_flag	= __PAGE_KERNEL_LARGE_EXEC, | 
|---|
| 245 | .kernpg_flag	= _KERNPG_TABLE_NOENC, | 
|---|
| 246 | }; | 
|---|
| 247 | unsigned long mstart, mend; | 
|---|
| 248 | int result; | 
|---|
| 249 | int i; | 
|---|
| 250 |  | 
|---|
| 251 | image->arch.pgd = alloc_pgt_page(data: image); | 
|---|
| 252 | if (!image->arch.pgd) | 
|---|
| 253 | return -ENOMEM; | 
|---|
| 254 |  | 
|---|
| 255 | if (cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT)) { | 
|---|
| 256 | info.page_flag   |= _PAGE_ENC; | 
|---|
| 257 | info.kernpg_flag |= _PAGE_ENC; | 
|---|
| 258 | } | 
|---|
| 259 |  | 
|---|
| 260 | if (direct_gbpages) | 
|---|
| 261 | info.direct_gbpages = true; | 
|---|
| 262 |  | 
|---|
| 263 | for (i = 0; i < nr_pfn_mapped; i++) { | 
|---|
| 264 | mstart = pfn_mapped[i].start << PAGE_SHIFT; | 
|---|
| 265 | mend   = pfn_mapped[i].end << PAGE_SHIFT; | 
|---|
| 266 |  | 
|---|
| 267 | result = kernel_ident_mapping_init(info: &info, pgd_page: image->arch.pgd, | 
|---|
| 268 | pstart: mstart, pend: mend); | 
|---|
| 269 | if (result) | 
|---|
| 270 | return result; | 
|---|
| 271 | } | 
|---|
| 272 |  | 
|---|
| 273 | /* | 
|---|
| 274 | * segments's mem ranges could be outside 0 ~ max_pfn, | 
|---|
| 275 | * for example when jump back to original kernel from kexeced kernel. | 
|---|
| 276 | * or first kernel is booted with user mem map, and second kernel | 
|---|
| 277 | * could be loaded out of that range. | 
|---|
| 278 | */ | 
|---|
| 279 | for (i = 0; i < image->nr_segments; i++) { | 
|---|
| 280 | mstart = image->segment[i].mem; | 
|---|
| 281 | mend   = mstart + image->segment[i].memsz; | 
|---|
| 282 |  | 
|---|
| 283 | result = kernel_ident_mapping_init(info: &info, pgd_page: image->arch.pgd, | 
|---|
| 284 | pstart: mstart, pend: mend); | 
|---|
| 285 |  | 
|---|
| 286 | if (result) | 
|---|
| 287 | return result; | 
|---|
| 288 | } | 
|---|
| 289 |  | 
|---|
| 290 | /* | 
|---|
| 291 | * Prepare EFI systab and ACPI tables for kexec kernel since they are | 
|---|
| 292 | * not covered by pfn_mapped. | 
|---|
| 293 | */ | 
|---|
| 294 | result = map_efi_systab(info: &info, level4p: image->arch.pgd); | 
|---|
| 295 | if (result) | 
|---|
| 296 | return result; | 
|---|
| 297 |  | 
|---|
| 298 | result = map_acpi_tables(info: &info, level4p: image->arch.pgd); | 
|---|
| 299 | if (result) | 
|---|
| 300 | return result; | 
|---|
| 301 |  | 
|---|
| 302 | result = map_mmio_serial(info: &info, level4p: image->arch.pgd); | 
|---|
| 303 | if (result) | 
|---|
| 304 | return result; | 
|---|
| 305 |  | 
|---|
| 306 | /* | 
|---|
| 307 | * This must be last because the intermediate page table pages it | 
|---|
| 308 | * allocates will not be control pages and may overlap the image. | 
|---|
| 309 | */ | 
|---|
| 310 | return init_transition_pgtable(image, pgd: image->arch.pgd, control_page); | 
|---|
| 311 | } | 
|---|
| 312 |  | 
|---|
| 313 | static void load_segments(void) | 
|---|
| 314 | { | 
|---|
| 315 | __asm__ __volatile__ ( | 
|---|
| 316 | "\tmovl %0,%%ds\n" | 
|---|
| 317 | "\tmovl %0,%%es\n" | 
|---|
| 318 | "\tmovl %0,%%ss\n" | 
|---|
| 319 | "\tmovl %0,%%fs\n" | 
|---|
| 320 | "\tmovl %0,%%gs\n" | 
|---|
| 321 | : : "a"(__KERNEL_DS) : "memory" | 
|---|
| 322 | ); | 
|---|
| 323 | } | 
|---|
| 324 |  | 
|---|
| 325 | static void prepare_debug_idt(unsigned long control_page, unsigned long vec_ofs) | 
|---|
| 326 | { | 
|---|
| 327 | gate_desc idtentry = { 0 }; | 
|---|
| 328 | int i; | 
|---|
| 329 |  | 
|---|
| 330 | idtentry.bits.p		= 1; | 
|---|
| 331 | idtentry.bits.type	= GATE_TRAP; | 
|---|
| 332 | idtentry.segment	= __KERNEL_CS; | 
|---|
| 333 | idtentry.offset_low	= (control_page & 0xFFFF) + vec_ofs; | 
|---|
| 334 | idtentry.offset_middle	= (control_page >> 16) & 0xFFFF; | 
|---|
| 335 | idtentry.offset_high	= control_page >> 32; | 
|---|
| 336 |  | 
|---|
| 337 | for (i = 0; i < 16; i++) { | 
|---|
| 338 | kexec_debug_idt[i] = idtentry; | 
|---|
| 339 | idtentry.offset_low += KEXEC_DEBUG_EXC_HANDLER_SIZE; | 
|---|
| 340 | } | 
|---|
| 341 | } | 
|---|
| 342 |  | 
|---|
| 343 | int machine_kexec_prepare(struct kimage *image) | 
|---|
| 344 | { | 
|---|
| 345 | void *control_page = page_address(image->control_code_page); | 
|---|
| 346 | unsigned long reloc_start = (unsigned long)__relocate_kernel_start; | 
|---|
| 347 | unsigned long reloc_end = (unsigned long)__relocate_kernel_end; | 
|---|
| 348 | int result; | 
|---|
| 349 |  | 
|---|
| 350 | /* | 
|---|
| 351 | * Some early TDX-capable platforms have an erratum.  A kernel | 
|---|
| 352 | * partial write (a write transaction of less than cacheline | 
|---|
| 353 | * lands at memory controller) to TDX private memory poisons that | 
|---|
| 354 | * memory, and a subsequent read triggers a machine check. | 
|---|
| 355 | * | 
|---|
| 356 | * On those platforms the old kernel must reset TDX private | 
|---|
| 357 | * memory before jumping to the new kernel otherwise the new | 
|---|
| 358 | * kernel may see unexpected machine check.  For simplicity | 
|---|
| 359 | * just fail kexec/kdump on those platforms. | 
|---|
| 360 | */ | 
|---|
| 361 | if (boot_cpu_has_bug(X86_BUG_TDX_PW_MCE)) { | 
|---|
| 362 | pr_info_once( "Not allowed on platform with tdx_pw_mce bug\n"); | 
|---|
| 363 | return -EOPNOTSUPP; | 
|---|
| 364 | } | 
|---|
| 365 |  | 
|---|
| 366 | /* Setup the identity mapped 64bit page table */ | 
|---|
| 367 | result = init_pgtable(image, __pa(control_page)); | 
|---|
| 368 | if (result) | 
|---|
| 369 | return result; | 
|---|
| 370 | kexec_va_control_page = (unsigned long)control_page; | 
|---|
| 371 | kexec_pa_table_page = (unsigned long)__pa(image->arch.pgd); | 
|---|
| 372 |  | 
|---|
| 373 | if (image->type == KEXEC_TYPE_DEFAULT) | 
|---|
| 374 | kexec_pa_swap_page = page_to_pfn(image->swap_page) << PAGE_SHIFT; | 
|---|
| 375 |  | 
|---|
| 376 | prepare_debug_idt(control_page: (unsigned long)__pa(control_page), | 
|---|
| 377 | vec_ofs: (unsigned long)kexec_debug_exc_vectors - reloc_start); | 
|---|
| 378 |  | 
|---|
| 379 | __memcpy(to: control_page, from: __relocate_kernel_start, len: reloc_end - reloc_start); | 
|---|
| 380 |  | 
|---|
| 381 | set_memory_rox(addr: (unsigned long)control_page, numpages: 1); | 
|---|
| 382 |  | 
|---|
| 383 | return 0; | 
|---|
| 384 | } | 
|---|
| 385 |  | 
|---|
| 386 | void machine_kexec_cleanup(struct kimage *image) | 
|---|
| 387 | { | 
|---|
| 388 | void *control_page = page_address(image->control_code_page); | 
|---|
| 389 |  | 
|---|
| 390 | set_memory_nx(addr: (unsigned long)control_page, numpages: 1); | 
|---|
| 391 | set_memory_rw(addr: (unsigned long)control_page, numpages: 1); | 
|---|
| 392 |  | 
|---|
| 393 | free_transition_pgtable(image); | 
|---|
| 394 | } | 
|---|
| 395 |  | 
|---|
| 396 | /* | 
|---|
| 397 | * Do not allocate memory (or fail in any way) in machine_kexec(). | 
|---|
| 398 | * We are past the point of no return, committed to rebooting now. | 
|---|
| 399 | */ | 
|---|
| 400 | void __nocfi machine_kexec(struct kimage *image) | 
|---|
| 401 | { | 
|---|
| 402 | unsigned long reloc_start = (unsigned long)__relocate_kernel_start; | 
|---|
| 403 | relocate_kernel_fn *relocate_kernel_ptr; | 
|---|
| 404 | unsigned int relocate_kernel_flags; | 
|---|
| 405 | int save_ftrace_enabled; | 
|---|
| 406 | void *control_page; | 
|---|
| 407 |  | 
|---|
| 408 | #ifdef CONFIG_KEXEC_JUMP | 
|---|
| 409 | if (image->preserve_context) | 
|---|
| 410 | save_processor_state(); | 
|---|
| 411 | #endif | 
|---|
| 412 |  | 
|---|
| 413 | save_ftrace_enabled = __ftrace_enabled_save(); | 
|---|
| 414 |  | 
|---|
| 415 | /* Interrupts aren't acceptable while we reboot */ | 
|---|
| 416 | local_irq_disable(); | 
|---|
| 417 | hw_breakpoint_disable(); | 
|---|
| 418 | cet_disable(); | 
|---|
| 419 |  | 
|---|
| 420 | if (image->preserve_context) { | 
|---|
| 421 | #ifdef CONFIG_X86_IO_APIC | 
|---|
| 422 | /* | 
|---|
| 423 | * We need to put APICs in legacy mode so that we can | 
|---|
| 424 | * get timer interrupts in second kernel. kexec/kdump | 
|---|
| 425 | * paths already have calls to restore_boot_irq_mode() | 
|---|
| 426 | * in one form or other. kexec jump path also need one. | 
|---|
| 427 | */ | 
|---|
| 428 | clear_IO_APIC(); | 
|---|
| 429 | restore_boot_irq_mode(); | 
|---|
| 430 | #endif | 
|---|
| 431 | } | 
|---|
| 432 |  | 
|---|
| 433 | control_page = page_address(image->control_code_page); | 
|---|
| 434 |  | 
|---|
| 435 | /* | 
|---|
| 436 | * Allow for the possibility that relocate_kernel might not be at | 
|---|
| 437 | * the very start of the page. | 
|---|
| 438 | */ | 
|---|
| 439 | relocate_kernel_ptr = control_page + (unsigned long)relocate_kernel - reloc_start; | 
|---|
| 440 |  | 
|---|
| 441 | relocate_kernel_flags = 0; | 
|---|
| 442 | if (image->preserve_context) | 
|---|
| 443 | relocate_kernel_flags |= RELOC_KERNEL_PRESERVE_CONTEXT; | 
|---|
| 444 |  | 
|---|
| 445 | /* | 
|---|
| 446 | * This must be done before load_segments() since it resets | 
|---|
| 447 | * GS to 0 and percpu data needs the correct GS to work. | 
|---|
| 448 | */ | 
|---|
| 449 | if (this_cpu_read(cache_state_incoherent)) | 
|---|
| 450 | relocate_kernel_flags |= RELOC_KERNEL_CACHE_INCOHERENT; | 
|---|
| 451 |  | 
|---|
| 452 | /* | 
|---|
| 453 | * The segment registers are funny things, they have both a | 
|---|
| 454 | * visible and an invisible part.  Whenever the visible part is | 
|---|
| 455 | * set to a specific selector, the invisible part is loaded | 
|---|
| 456 | * with from a table in memory.  At no other time is the | 
|---|
| 457 | * descriptor table in memory accessed. | 
|---|
| 458 | * | 
|---|
| 459 | * Take advantage of this here by force loading the segments, | 
|---|
| 460 | * before the GDT is zapped with an invalid value. | 
|---|
| 461 | * | 
|---|
| 462 | * load_segments() resets GS to 0.  Don't make any function call | 
|---|
| 463 | * after here since call depth tracking uses percpu variables to | 
|---|
| 464 | * operate (relocate_kernel() is explicitly ignored by call depth | 
|---|
| 465 | * tracking). | 
|---|
| 466 | */ | 
|---|
| 467 | load_segments(); | 
|---|
| 468 |  | 
|---|
| 469 | /* now call it */ | 
|---|
| 470 | image->start = relocate_kernel_ptr((unsigned long)image->head, | 
|---|
| 471 | virt_to_phys(address: control_page), | 
|---|
| 472 | image->start, | 
|---|
| 473 | relocate_kernel_flags); | 
|---|
| 474 |  | 
|---|
| 475 | #ifdef CONFIG_KEXEC_JUMP | 
|---|
| 476 | if (image->preserve_context) | 
|---|
| 477 | restore_processor_state(); | 
|---|
| 478 | #endif | 
|---|
| 479 |  | 
|---|
| 480 | __ftrace_enabled_restore(enabled: save_ftrace_enabled); | 
|---|
| 481 | } | 
|---|
| 482 | /* | 
|---|
| 483 | * Handover to the next kernel, no CFI concern. | 
|---|
| 484 | */ | 
|---|
| 485 | ANNOTATE_NOCFI_SYM(machine_kexec); | 
|---|
| 486 |  | 
|---|
| 487 | /* arch-dependent functionality related to kexec file-based syscall */ | 
|---|
| 488 |  | 
|---|
| 489 | #ifdef CONFIG_KEXEC_FILE | 
|---|
| 490 | /* | 
|---|
| 491 | * Apply purgatory relocations. | 
|---|
| 492 | * | 
|---|
| 493 | * @pi:		Purgatory to be relocated. | 
|---|
| 494 | * @section:	Section relocations applying to. | 
|---|
| 495 | * @relsec:	Section containing RELAs. | 
|---|
| 496 | * @symtabsec:	Corresponding symtab. | 
|---|
| 497 | * | 
|---|
| 498 | * TODO: Some of the code belongs to generic code. Move that in kexec.c. | 
|---|
| 499 | */ | 
|---|
| 500 | int arch_kexec_apply_relocations_add(struct purgatory_info *pi, | 
|---|
| 501 | Elf_Shdr *section, const Elf_Shdr *relsec, | 
|---|
| 502 | const Elf_Shdr *symtabsec) | 
|---|
| 503 | { | 
|---|
| 504 | unsigned int i; | 
|---|
| 505 | Elf64_Rela *rel; | 
|---|
| 506 | Elf64_Sym *sym; | 
|---|
| 507 | void *location; | 
|---|
| 508 | unsigned long address, sec_base, value; | 
|---|
| 509 | const char *strtab, *name, *shstrtab; | 
|---|
| 510 | const Elf_Shdr *sechdrs; | 
|---|
| 511 |  | 
|---|
| 512 | /* String & section header string table */ | 
|---|
| 513 | sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff; | 
|---|
| 514 | strtab = (char *)pi->ehdr + sechdrs[symtabsec->sh_link].sh_offset; | 
|---|
| 515 | shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset; | 
|---|
| 516 |  | 
|---|
| 517 | rel = (void *)pi->ehdr + relsec->sh_offset; | 
|---|
| 518 |  | 
|---|
| 519 | pr_debug( "Applying relocate section %s to %u\n", | 
|---|
| 520 | shstrtab + relsec->sh_name, relsec->sh_info); | 
|---|
| 521 |  | 
|---|
| 522 | for (i = 0; i < relsec->sh_size / sizeof(*rel); i++) { | 
|---|
| 523 |  | 
|---|
| 524 | /* | 
|---|
| 525 | * rel[i].r_offset contains byte offset from beginning | 
|---|
| 526 | * of section to the storage unit affected. | 
|---|
| 527 | * | 
|---|
| 528 | * This is location to update. This is temporary buffer | 
|---|
| 529 | * where section is currently loaded. This will finally be | 
|---|
| 530 | * loaded to a different address later, pointed to by | 
|---|
| 531 | * ->sh_addr. kexec takes care of moving it | 
|---|
| 532 | *  (kexec_load_segment()). | 
|---|
| 533 | */ | 
|---|
| 534 | location = pi->purgatory_buf; | 
|---|
| 535 | location += section->sh_offset; | 
|---|
| 536 | location += rel[i].r_offset; | 
|---|
| 537 |  | 
|---|
| 538 | /* Final address of the location */ | 
|---|
| 539 | address = section->sh_addr + rel[i].r_offset; | 
|---|
| 540 |  | 
|---|
| 541 | /* | 
|---|
| 542 | * rel[i].r_info contains information about symbol table index | 
|---|
| 543 | * w.r.t which relocation must be made and type of relocation | 
|---|
| 544 | * to apply. ELF64_R_SYM() and ELF64_R_TYPE() macros get | 
|---|
| 545 | * these respectively. | 
|---|
| 546 | */ | 
|---|
| 547 | sym = (void *)pi->ehdr + symtabsec->sh_offset; | 
|---|
| 548 | sym += ELF64_R_SYM(rel[i].r_info); | 
|---|
| 549 |  | 
|---|
| 550 | if (sym->st_name) | 
|---|
| 551 | name = strtab + sym->st_name; | 
|---|
| 552 | else | 
|---|
| 553 | name = shstrtab + sechdrs[sym->st_shndx].sh_name; | 
|---|
| 554 |  | 
|---|
| 555 | pr_debug( "Symbol: %s info: %02x shndx: %02x value=%llx size: %llx\n", | 
|---|
| 556 | name, sym->st_info, sym->st_shndx, sym->st_value, | 
|---|
| 557 | sym->st_size); | 
|---|
| 558 |  | 
|---|
| 559 | if (sym->st_shndx == SHN_UNDEF) { | 
|---|
| 560 | pr_err( "Undefined symbol: %s\n", name); | 
|---|
| 561 | return -ENOEXEC; | 
|---|
| 562 | } | 
|---|
| 563 |  | 
|---|
| 564 | if (sym->st_shndx == SHN_COMMON) { | 
|---|
| 565 | pr_err( "symbol '%s' in common section\n", name); | 
|---|
| 566 | return -ENOEXEC; | 
|---|
| 567 | } | 
|---|
| 568 |  | 
|---|
| 569 | if (sym->st_shndx == SHN_ABS) | 
|---|
| 570 | sec_base = 0; | 
|---|
| 571 | else if (sym->st_shndx >= pi->ehdr->e_shnum) { | 
|---|
| 572 | pr_err( "Invalid section %d for symbol %s\n", | 
|---|
| 573 | sym->st_shndx, name); | 
|---|
| 574 | return -ENOEXEC; | 
|---|
| 575 | } else | 
|---|
| 576 | sec_base = pi->sechdrs[sym->st_shndx].sh_addr; | 
|---|
| 577 |  | 
|---|
| 578 | value = sym->st_value; | 
|---|
| 579 | value += sec_base; | 
|---|
| 580 | value += rel[i].r_addend; | 
|---|
| 581 |  | 
|---|
| 582 | switch (ELF64_R_TYPE(rel[i].r_info)) { | 
|---|
| 583 | case R_X86_64_NONE: | 
|---|
| 584 | break; | 
|---|
| 585 | case R_X86_64_64: | 
|---|
| 586 | *(u64 *)location = value; | 
|---|
| 587 | break; | 
|---|
| 588 | case R_X86_64_32: | 
|---|
| 589 | *(u32 *)location = value; | 
|---|
| 590 | if (value != *(u32 *)location) | 
|---|
| 591 | goto overflow; | 
|---|
| 592 | break; | 
|---|
| 593 | case R_X86_64_32S: | 
|---|
| 594 | *(s32 *)location = value; | 
|---|
| 595 | if ((s64)value != *(s32 *)location) | 
|---|
| 596 | goto overflow; | 
|---|
| 597 | break; | 
|---|
| 598 | case R_X86_64_PC32: | 
|---|
| 599 | case R_X86_64_PLT32: | 
|---|
| 600 | value -= (u64)address; | 
|---|
| 601 | *(u32 *)location = value; | 
|---|
| 602 | break; | 
|---|
| 603 | default: | 
|---|
| 604 | pr_err( "Unknown rela relocation: %llu\n", | 
|---|
| 605 | ELF64_R_TYPE(rel[i].r_info)); | 
|---|
| 606 | return -ENOEXEC; | 
|---|
| 607 | } | 
|---|
| 608 | } | 
|---|
| 609 | return 0; | 
|---|
| 610 |  | 
|---|
| 611 | overflow: | 
|---|
| 612 | pr_err( "Overflow in relocation type %d value 0x%lx\n", | 
|---|
| 613 | (int)ELF64_R_TYPE(rel[i].r_info), value); | 
|---|
| 614 | return -ENOEXEC; | 
|---|
| 615 | } | 
|---|
| 616 |  | 
|---|
| 617 | int arch_kimage_file_post_load_cleanup(struct kimage *image) | 
|---|
| 618 | { | 
|---|
| 619 | vfree(image->elf_headers); | 
|---|
| 620 | image->elf_headers = NULL; | 
|---|
| 621 | image->elf_headers_sz = 0; | 
|---|
| 622 |  | 
|---|
| 623 | return kexec_image_post_load_cleanup_default(image); | 
|---|
| 624 | } | 
|---|
| 625 | #endif /* CONFIG_KEXEC_FILE */ | 
|---|
| 626 |  | 
|---|
| 627 | #ifdef CONFIG_CRASH_DUMP | 
|---|
| 628 |  | 
|---|
| 629 | static int | 
|---|
| 630 | kexec_mark_range(unsigned long start, unsigned long end, bool protect) | 
|---|
| 631 | { | 
|---|
| 632 | struct page *page; | 
|---|
| 633 | unsigned int nr_pages; | 
|---|
| 634 |  | 
|---|
| 635 | /* | 
|---|
| 636 | * For physical range: [start, end]. We must skip the unassigned | 
|---|
| 637 | * crashk resource with zero-valued "end" member. | 
|---|
| 638 | */ | 
|---|
| 639 | if (!end || start > end) | 
|---|
| 640 | return 0; | 
|---|
| 641 |  | 
|---|
| 642 | page = pfn_to_page(start >> PAGE_SHIFT); | 
|---|
| 643 | nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1; | 
|---|
| 644 | if (protect) | 
|---|
| 645 | return set_pages_ro(page, numpages: nr_pages); | 
|---|
| 646 | else | 
|---|
| 647 | return set_pages_rw(page, numpages: nr_pages); | 
|---|
| 648 | } | 
|---|
| 649 |  | 
|---|
| 650 | static void kexec_mark_crashkres(bool protect) | 
|---|
| 651 | { | 
|---|
| 652 | unsigned long control; | 
|---|
| 653 |  | 
|---|
| 654 | kexec_mark_range(start: crashk_low_res.start, end: crashk_low_res.end, protect); | 
|---|
| 655 |  | 
|---|
| 656 | /* Don't touch the control code page used in crash_kexec().*/ | 
|---|
| 657 | control = PFN_PHYS(page_to_pfn(kexec_crash_image->control_code_page)); | 
|---|
| 658 | kexec_mark_range(start: crashk_res.start, end: control - 1, protect); | 
|---|
| 659 | control += KEXEC_CONTROL_PAGE_SIZE; | 
|---|
| 660 | kexec_mark_range(start: control, end: crashk_res.end, protect); | 
|---|
| 661 | } | 
|---|
| 662 |  | 
|---|
| 663 | /* make the memory storing dm crypt keys in/accessible */ | 
|---|
| 664 | static void kexec_mark_dm_crypt_keys(bool protect) | 
|---|
| 665 | { | 
|---|
| 666 | unsigned long start_paddr, end_paddr; | 
|---|
| 667 | unsigned int nr_pages; | 
|---|
| 668 |  | 
|---|
| 669 | if (kexec_crash_image->dm_crypt_keys_addr) { | 
|---|
| 670 | start_paddr = kexec_crash_image->dm_crypt_keys_addr; | 
|---|
| 671 | end_paddr = start_paddr + kexec_crash_image->dm_crypt_keys_sz - 1; | 
|---|
| 672 | nr_pages = (PAGE_ALIGN(end_paddr) - PAGE_ALIGN_DOWN(start_paddr))/PAGE_SIZE; | 
|---|
| 673 | if (protect) | 
|---|
| 674 | set_memory_np(addr: (unsigned long)phys_to_virt(address: start_paddr), numpages: nr_pages); | 
|---|
| 675 | else | 
|---|
| 676 | __set_memory_prot( | 
|---|
| 677 | addr: (unsigned long)phys_to_virt(address: start_paddr), | 
|---|
| 678 | numpages: nr_pages, | 
|---|
| 679 | __pgprot(_PAGE_PRESENT | _PAGE_NX | _PAGE_RW)); | 
|---|
| 680 | } | 
|---|
| 681 | } | 
|---|
| 682 |  | 
|---|
| 683 | void arch_kexec_protect_crashkres(void) | 
|---|
| 684 | { | 
|---|
| 685 | kexec_mark_crashkres(protect: true); | 
|---|
| 686 | kexec_mark_dm_crypt_keys(protect: true); | 
|---|
| 687 | } | 
|---|
| 688 |  | 
|---|
| 689 | void arch_kexec_unprotect_crashkres(void) | 
|---|
| 690 | { | 
|---|
| 691 | kexec_mark_dm_crypt_keys(protect: false); | 
|---|
| 692 | kexec_mark_crashkres(protect: false); | 
|---|
| 693 | } | 
|---|
| 694 | #endif | 
|---|
| 695 |  | 
|---|
| 696 | /* | 
|---|
| 697 | * During a traditional boot under SME, SME will encrypt the kernel, | 
|---|
| 698 | * so the SME kexec kernel also needs to be un-encrypted in order to | 
|---|
| 699 | * replicate a normal SME boot. | 
|---|
| 700 | * | 
|---|
| 701 | * During a traditional boot under SEV, the kernel has already been | 
|---|
| 702 | * loaded encrypted, so the SEV kexec kernel needs to be encrypted in | 
|---|
| 703 | * order to replicate a normal SEV boot. | 
|---|
| 704 | */ | 
|---|
| 705 | int arch_kexec_post_alloc_pages(void *vaddr, unsigned int pages, gfp_t gfp) | 
|---|
| 706 | { | 
|---|
| 707 | if (!cc_platform_has(attr: CC_ATTR_HOST_MEM_ENCRYPT)) | 
|---|
| 708 | return 0; | 
|---|
| 709 |  | 
|---|
| 710 | /* | 
|---|
| 711 | * If host memory encryption is active we need to be sure that kexec | 
|---|
| 712 | * pages are not encrypted because when we boot to the new kernel the | 
|---|
| 713 | * pages won't be accessed encrypted (initially). | 
|---|
| 714 | */ | 
|---|
| 715 | return set_memory_decrypted(addr: (unsigned long)vaddr, numpages: pages); | 
|---|
| 716 | } | 
|---|
| 717 |  | 
|---|
| 718 | void arch_kexec_pre_free_pages(void *vaddr, unsigned int pages) | 
|---|
| 719 | { | 
|---|
| 720 | if (!cc_platform_has(attr: CC_ATTR_HOST_MEM_ENCRYPT)) | 
|---|
| 721 | return; | 
|---|
| 722 |  | 
|---|
| 723 | /* | 
|---|
| 724 | * If host memory encryption is active we need to reset the pages back | 
|---|
| 725 | * to being an encrypted mapping before freeing them. | 
|---|
| 726 | */ | 
|---|
| 727 | set_memory_encrypted(addr: (unsigned long)vaddr, numpages: pages); | 
|---|
| 728 | } | 
|---|
| 729 |  | 
|---|