1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * ld script for the x86 kernel 4 * 5 * Historic 32-bit version written by Martin Mares <mj@atrey.karlin.mff.cuni.cz> 6 * 7 * Modernisation, unification and other changes and fixes: 8 * Copyright (C) 2007-2009 Sam Ravnborg <sam@ravnborg.org> 9 * 10 * 11 * Don't define absolute symbols until and unless you know that symbol 12 * value is should remain constant even if kernel image is relocated 13 * at run time. Absolute symbols are not relocated. If symbol value should 14 * change if kernel is relocated, make the symbol section relative and 15 * put it inside the section definition. 16 */ 17 18#define LOAD_OFFSET __START_KERNEL_map 19 20#define RUNTIME_DISCARD_EXIT 21#define EMITS_PT_NOTE 22#define RO_EXCEPTION_TABLE_ALIGN 16 23 24#include <asm-generic/vmlinux.lds.h> 25#include <asm/asm-offsets.h> 26#include <asm/thread_info.h> 27#include <asm/page_types.h> 28#include <asm/orc_lookup.h> 29#include <asm/cache.h> 30#include <asm/boot.h> 31#include <asm/kexec.h> 32 33#undef i386 /* in case the preprocessor is a 32bit one */ 34 35OUTPUT_FORMAT(CONFIG_OUTPUT_FORMAT) 36 37#ifdef CONFIG_X86_32 38OUTPUT_ARCH(i386) 39ENTRY(phys_startup_32) 40#else 41OUTPUT_ARCH(i386:x86-64) 42ENTRY(phys_startup_64) 43#endif 44 45jiffies = jiffies_64; 46const_current_task = current_task; 47const_cpu_current_top_of_stack = cpu_current_top_of_stack; 48 49#if defined(CONFIG_X86_64) 50/* 51 * On 64-bit, align RODATA to 2MB so we retain large page mappings for 52 * boundaries spanning kernel text, rodata and data sections. 53 * 54 * However, kernel identity mappings will have different RWX permissions 55 * to the pages mapping to text and to the pages padding (which are freed) the 56 * text section. Hence kernel identity mappings will be broken to smaller 57 * pages. For 64-bit, kernel text and kernel identity mappings are different, 58 * so we can enable protection checks as well as retain 2MB large page 59 * mappings for kernel text. 60 */ 61#define X86_ALIGN_RODATA_BEGIN . = ALIGN(HPAGE_SIZE); 62 63#define X86_ALIGN_RODATA_END \ 64 . = ALIGN(HPAGE_SIZE); \ 65 __end_rodata_hpage_align = .; \ 66 __end_rodata_aligned = .; 67 68#define ALIGN_ENTRY_TEXT_BEGIN . = ALIGN(PMD_SIZE); 69#define ALIGN_ENTRY_TEXT_END . = ALIGN(PMD_SIZE); 70 71/* 72 * This section contains data which will be mapped as decrypted. Memory 73 * encryption operates on a page basis. Make this section PMD-aligned 74 * to avoid splitting the pages while mapping the section early. 75 * 76 * Note: We use a separate section so that only this section gets 77 * decrypted to avoid exposing more than we wish. 78 */ 79#define BSS_DECRYPTED \ 80 . = ALIGN(PMD_SIZE); \ 81 __start_bss_decrypted = .; \ 82 *(.bss..decrypted); \ 83 . = ALIGN(PAGE_SIZE); \ 84 __start_bss_decrypted_unused = .; \ 85 . = ALIGN(PMD_SIZE); \ 86 __end_bss_decrypted = .; \ 87 88#else 89 90#define X86_ALIGN_RODATA_BEGIN 91#define X86_ALIGN_RODATA_END \ 92 . = ALIGN(PAGE_SIZE); \ 93 __end_rodata_aligned = .; 94 95#define ALIGN_ENTRY_TEXT_BEGIN 96#define ALIGN_ENTRY_TEXT_END 97#define BSS_DECRYPTED 98 99#endif 100#if defined(CONFIG_X86_64) && defined(CONFIG_KEXEC_CORE) 101#define KEXEC_RELOCATE_KERNEL \ 102 . = ALIGN(0x100); \ 103 __relocate_kernel_start = .; \ 104 *(.text..relocate_kernel); \ 105 *(.data..relocate_kernel); \ 106 __relocate_kernel_end = .; 107 108ASSERT(__relocate_kernel_end - __relocate_kernel_start <= KEXEC_CONTROL_CODE_MAX_SIZE, 109 "relocate_kernel code too large!") 110#else 111#define KEXEC_RELOCATE_KERNEL 112#endif 113PHDRS { 114 text PT_LOAD FLAGS(5); /* R_E */ 115 data PT_LOAD FLAGS(6); /* RW_ */ 116 note PT_NOTE FLAGS(0); /* ___ */ 117} 118 119SECTIONS 120{ 121 . = __START_KERNEL; 122#ifdef CONFIG_X86_32 123 phys_startup_32 = ABSOLUTE(startup_32 - LOAD_OFFSET); 124#else 125 phys_startup_64 = ABSOLUTE(startup_64 - LOAD_OFFSET); 126#endif 127 128 /* Text and read-only data */ 129 .text : AT(ADDR(.text) - LOAD_OFFSET) { 130 _text = .; 131 _stext = .; 132 ALIGN_ENTRY_TEXT_BEGIN 133 *(.text..__x86.rethunk_untrain) 134 ENTRY_TEXT 135 136#ifdef CONFIG_MITIGATION_SRSO 137 /* 138 * See the comment above srso_alias_untrain_ret()'s 139 * definition. 140 */ 141 . = srso_alias_untrain_ret | (1 << 2) | (1 << 8) | (1 << 14) | (1 << 20); 142 *(.text..__x86.rethunk_safe) 143#endif 144 ALIGN_ENTRY_TEXT_END 145 146 TEXT_TEXT 147 SCHED_TEXT 148 LOCK_TEXT 149 KPROBES_TEXT 150 SOFTIRQENTRY_TEXT 151#ifdef CONFIG_MITIGATION_RETPOLINE 152 *(.text..__x86.indirect_thunk) 153 *(.text..__x86.return_thunk) 154#endif 155 STATIC_CALL_TEXT 156 *(.gnu.warning) 157 158 } :text = 0xcccccccc 159 160 /* bootstrapping code */ 161 .head.text : AT(ADDR(.head.text) - LOAD_OFFSET) { 162 HEAD_TEXT 163 } :text = 0xcccccccc 164 165 /* End of text section, which should occupy whole number of pages */ 166 _etext = .; 167 . = ALIGN(PAGE_SIZE); 168 169 X86_ALIGN_RODATA_BEGIN 170 RO_DATA(PAGE_SIZE) 171 X86_ALIGN_RODATA_END 172 173 /* Data */ 174 .data : AT(ADDR(.data) - LOAD_OFFSET) { 175 /* Start of data section */ 176 _sdata = .; 177 178 /* init_task */ 179 INIT_TASK_DATA(THREAD_SIZE) 180 181 /* equivalent to task_pt_regs(&init_task) */ 182 __top_init_kernel_stack = __end_init_stack - TOP_OF_KERNEL_STACK_PADDING - PTREGS_SIZE; 183 184#ifdef CONFIG_X86_32 185 /* 32 bit has nosave before _edata */ 186 NOSAVE_DATA 187#endif 188 189 PAGE_ALIGNED_DATA(PAGE_SIZE) 190 191 CACHE_HOT_DATA(L1_CACHE_BYTES) 192 193 CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES) 194 195 DATA_DATA 196 CONSTRUCTORS 197 KEXEC_RELOCATE_KERNEL 198 199 /* rarely changed data like cpu maps */ 200 READ_MOSTLY_DATA(INTERNODE_CACHE_BYTES) 201 202 /* End of data section */ 203 _edata = .; 204 } :data 205 206 BUG_TABLE 207 208 ORC_UNWIND_TABLE 209 210 /* Init code and data - will be freed after init */ 211 . = ALIGN(PAGE_SIZE); 212 .init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) { 213 __init_begin = .; /* paired with __init_end */ 214 } 215 216 INIT_TEXT_SECTION(PAGE_SIZE) 217 218 /* 219 * Section for code used exclusively before alternatives are run. All 220 * references to such code must be patched out by alternatives, normally 221 * by using X86_FEATURE_ALWAYS CPU feature bit. 222 * 223 * See static_cpu_has() for an example. 224 */ 225 .altinstr_aux : AT(ADDR(.altinstr_aux) - LOAD_OFFSET) { 226 *(.altinstr_aux) 227 } 228 229 INIT_DATA_SECTION(16) 230 231 .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) { 232 __x86_cpu_dev_start = .; 233 *(.x86_cpu_dev.init) 234 __x86_cpu_dev_end = .; 235 } 236 237#ifdef CONFIG_X86_INTEL_MID 238 .x86_intel_mid_dev.init : AT(ADDR(.x86_intel_mid_dev.init) - \ 239 LOAD_OFFSET) { 240 __x86_intel_mid_dev_start = .; 241 *(.x86_intel_mid_dev.init) 242 __x86_intel_mid_dev_end = .; 243 } 244#endif 245 246#ifdef CONFIG_MITIGATION_RETPOLINE 247 /* 248 * List of instructions that call/jmp/jcc to retpoline thunks 249 * __x86_indirect_thunk_*(). These instructions can be patched along 250 * with alternatives, after which the section can be freed. 251 */ 252 . = ALIGN(8); 253 .retpoline_sites : AT(ADDR(.retpoline_sites) - LOAD_OFFSET) { 254 __retpoline_sites = .; 255 *(.retpoline_sites) 256 __retpoline_sites_end = .; 257 } 258 259 . = ALIGN(8); 260 .return_sites : AT(ADDR(.return_sites) - LOAD_OFFSET) { 261 __return_sites = .; 262 *(.return_sites) 263 __return_sites_end = .; 264 } 265 266 . = ALIGN(8); 267 .call_sites : AT(ADDR(.call_sites) - LOAD_OFFSET) { 268 __call_sites = .; 269 *(.call_sites) 270 __call_sites_end = .; 271 } 272#endif 273 274#ifdef CONFIG_X86_KERNEL_IBT 275 . = ALIGN(8); 276 .ibt_endbr_seal : AT(ADDR(.ibt_endbr_seal) - LOAD_OFFSET) { 277 __ibt_endbr_seal = .; 278 *(.ibt_endbr_seal) 279 __ibt_endbr_seal_end = .; 280 } 281#endif 282 283#ifdef CONFIG_FINEIBT 284 . = ALIGN(8); 285 .cfi_sites : AT(ADDR(.cfi_sites) - LOAD_OFFSET) { 286 __cfi_sites = .; 287 *(.cfi_sites) 288 __cfi_sites_end = .; 289 } 290#endif 291 292 /* 293 * struct alt_inst entries. From the header (alternative.h): 294 * "Alternative instructions for different CPU types or capabilities" 295 * Think locking instructions on spinlocks. 296 */ 297 . = ALIGN(8); 298 .altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) { 299 __alt_instructions = .; 300 *(.altinstructions) 301 __alt_instructions_end = .; 302 } 303 304 /* 305 * And here are the replacement instructions. The linker sticks 306 * them as binary blobs. The .altinstructions has enough data to 307 * get the address and the length of them to patch the kernel safely. 308 */ 309 .altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) { 310 *(.altinstr_replacement) 311 } 312 313 . = ALIGN(8); 314 .apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) { 315 __apicdrivers = .; 316 *(.apicdrivers); 317 __apicdrivers_end = .; 318 } 319 320 . = ALIGN(8); 321 /* 322 * .exit.text is discarded at runtime, not link time, to deal with 323 * references from .altinstructions 324 */ 325 .exit.text : AT(ADDR(.exit.text) - LOAD_OFFSET) { 326 EXIT_TEXT 327 } 328 329 .exit.data : AT(ADDR(.exit.data) - LOAD_OFFSET) { 330 EXIT_DATA 331 } 332 333 PERCPU_SECTION(L1_CACHE_BYTES) 334 ASSERT(__per_cpu_hot_end - __per_cpu_hot_start <= 64, "percpu cache hot data too large") 335 336 RUNTIME_CONST_VARIABLES 337 RUNTIME_CONST(ptr, USER_PTR_MAX) 338 339 . = ALIGN(PAGE_SIZE); 340 341 /* freed after init ends here */ 342 .init.end : AT(ADDR(.init.end) - LOAD_OFFSET) { 343 __init_end = .; 344 } 345 346 /* 347 * smp_locks might be freed after init 348 * start/end must be page aligned 349 */ 350 . = ALIGN(PAGE_SIZE); 351 .smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) { 352 __smp_locks = .; 353 *(.smp_locks) 354 . = ALIGN(PAGE_SIZE); 355 __smp_locks_end = .; 356 } 357 358#ifdef CONFIG_X86_64 359 .data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) { 360 NOSAVE_DATA 361 } 362#endif 363 364 /* BSS */ 365 . = ALIGN(PAGE_SIZE); 366 .bss : AT(ADDR(.bss) - LOAD_OFFSET) { 367 __bss_start = .; 368 *(.bss..page_aligned) 369 . = ALIGN(PAGE_SIZE); 370 *(BSS_MAIN) 371 BSS_DECRYPTED 372 . = ALIGN(PAGE_SIZE); 373 __bss_stop = .; 374 } 375 376 /* 377 * The memory occupied from _text to here, __end_of_kernel_reserve, is 378 * automatically reserved in setup_arch(). Anything after here must be 379 * explicitly reserved using memblock_reserve() or it will be discarded 380 * and treated as available memory. 381 */ 382 __end_of_kernel_reserve = .; 383 384 . = ALIGN(PAGE_SIZE); 385 .brk : AT(ADDR(.brk) - LOAD_OFFSET) { 386 __brk_base = .; 387 . += 64 * 1024; /* 64k alignment slop space */ 388 *(.bss..brk) /* areas brk users have reserved */ 389 __brk_limit = .; 390 } 391 392 . = ALIGN(PAGE_SIZE); /* keep VO_INIT_SIZE page aligned */ 393 _end = .; 394 395#ifdef CONFIG_AMD_MEM_ENCRYPT 396 /* 397 * Early scratch/workarea section: Lives outside of the kernel proper 398 * (_text - _end). 399 * 400 * Resides after _end because even though the .brk section is after 401 * __end_of_kernel_reserve, the .brk section is later reserved as a 402 * part of the kernel. Since it is located after __end_of_kernel_reserve 403 * it will be discarded and become part of the available memory. As 404 * such, it can only be used by very early boot code and must not be 405 * needed afterwards. 406 * 407 * Currently used by SME for performing in-place encryption of the 408 * kernel during boot. Resides on a 2MB boundary to simplify the 409 * pagetable setup used for SME in-place encryption. 410 */ 411 . = ALIGN(HPAGE_SIZE); 412 .init.scratch : AT(ADDR(.init.scratch) - LOAD_OFFSET) { 413 __init_scratch_begin = .; 414 *(.init.scratch) 415 . = ALIGN(HPAGE_SIZE); 416 __init_scratch_end = .; 417 } 418#endif 419 420 STABS_DEBUG 421 DWARF_DEBUG 422#ifdef CONFIG_PROPELLER_CLANG 423 .llvm_bb_addr_map : { *(.llvm_bb_addr_map) } 424#endif 425 426 ELF_DETAILS 427 428 DISCARDS 429 430 /* 431 * Make sure that the .got.plt is either completely empty or it 432 * contains only the lazy dispatch entries. 433 */ 434 .got.plt (INFO) : { *(.got.plt) } 435 ASSERT(SIZEOF(.got.plt) == 0 || 436#ifdef CONFIG_X86_64 437 SIZEOF(.got.plt) == 0x18, 438#else 439 SIZEOF(.got.plt) == 0xc, 440#endif 441 "Unexpected GOT/PLT entries detected!") 442 443 /* 444 * Sections that should stay zero sized, which is safer to 445 * explicitly check instead of blindly discarding. 446 */ 447 .got : { 448 *(.got) *(.igot.*) 449 } 450 ASSERT(SIZEOF(.got) == 0, "Unexpected GOT entries detected!") 451 452 .plt : { 453 *(.plt) *(.plt.*) *(.iplt) 454 } 455 ASSERT(SIZEOF(.plt) == 0, "Unexpected run-time procedure linkages detected!") 456 457 .rel.dyn : { 458 *(.rel.*) *(.rel_*) 459 } 460 ASSERT(SIZEOF(.rel.dyn) == 0, "Unexpected run-time relocations (.rel) detected!") 461 462 .rela.dyn : { 463 *(.rela.*) *(.rela_*) 464 } 465 ASSERT(SIZEOF(.rela.dyn) == 0, "Unexpected run-time relocations (.rela) detected!") 466} 467 468/* 469 * COMPILE_TEST kernels can be large - CONFIG_KASAN, for example, can cause 470 * this. Let's assume that nobody will be running a COMPILE_TEST kernel and 471 * let's assert that fuller build coverage is more valuable than being able to 472 * run a COMPILE_TEST kernel. 473 */ 474#ifndef CONFIG_COMPILE_TEST 475/* 476 * The ASSERT() sync to . is intentional, for binutils 2.14 compatibility: 477 */ 478. = ASSERT((_end - LOAD_OFFSET <= KERNEL_IMAGE_SIZE), 479 "kernel image bigger than KERNEL_IMAGE_SIZE"); 480#endif 481 482/* needed for Clang - see arch/x86/entry/entry.S */ 483PROVIDE(__ref_stack_chk_guard = __stack_chk_guard); 484 485#ifdef CONFIG_X86_64 486 487#ifdef CONFIG_MITIGATION_UNRET_ENTRY 488. = ASSERT((retbleed_return_thunk & 0x3f) == 0, "retbleed_return_thunk not cacheline-aligned"); 489#endif 490 491#ifdef CONFIG_MITIGATION_SRSO 492. = ASSERT((srso_safe_ret & 0x3f) == 0, "srso_safe_ret not cacheline-aligned"); 493/* 494 * GNU ld cannot do XOR until 2.41. 495 * https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=f6f78318fca803c4907fb8d7f6ded8295f1947b1 496 * 497 * LLVM lld cannot do XOR until lld-17. 498 * https://github.com/llvm/llvm-project/commit/fae96104d4378166cbe5c875ef8ed808a356f3fb 499 * 500 * Instead do: (A | B) - (A & B) in order to compute the XOR 501 * of the two function addresses: 502 */ 503. = ASSERT(((ABSOLUTE(srso_alias_untrain_ret) | srso_alias_safe_ret) - 504 (ABSOLUTE(srso_alias_untrain_ret) & srso_alias_safe_ret)) == ((1 << 2) | (1 << 8) | (1 << 14) | (1 << 20)), 505 "SRSO function pair won't alias"); 506#endif 507 508#if defined(CONFIG_MITIGATION_ITS) && !defined(CONFIG_DEBUG_FORCE_FUNCTION_ALIGN_64B) 509. = ASSERT(__x86_indirect_its_thunk_rax & 0x20, "__x86_indirect_thunk_rax not in second half of cacheline"); 510. = ASSERT(((__x86_indirect_its_thunk_rcx - __x86_indirect_its_thunk_rax) % 64) == 0, "Indirect thunks are not cacheline apart"); 511. = ASSERT(__x86_indirect_its_thunk_array == __x86_indirect_its_thunk_rax, "Gap in ITS thunk array"); 512#endif 513 514#if defined(CONFIG_MITIGATION_ITS) && !defined(CONFIG_DEBUG_FORCE_FUNCTION_ALIGN_64B) 515. = ASSERT(its_return_thunk & 0x20, "its_return_thunk not in second half of cacheline"); 516#endif 517 518#endif /* CONFIG_X86_64 */ 519 520/* 521 * The symbols below are referenced using relative relocations in the 522 * respective ELF notes. This produces build time constants that the 523 * linker will never mark as relocatable. (Using just ABSOLUTE() is not 524 * sufficient for that). 525 */ 526#ifdef CONFIG_XEN_PV 527xen_elfnote_entry_value = 528 ABSOLUTE(xen_elfnote_entry) + ABSOLUTE(startup_xen); 529#endif 530#ifdef CONFIG_PVH 531xen_elfnote_phys32_entry_value = 532 ABSOLUTE(xen_elfnote_phys32_entry) + ABSOLUTE(pvh_start_xen - LOAD_OFFSET); 533#endif 534