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_pcpu_hot = pcpu_hot; 47 48#if defined(CONFIG_X86_64) 49/* 50 * On 64-bit, align RODATA to 2MB so we retain large page mappings for 51 * boundaries spanning kernel text, rodata and data sections. 52 * 53 * However, kernel identity mappings will have different RWX permissions 54 * to the pages mapping to text and to the pages padding (which are freed) the 55 * text section. Hence kernel identity mappings will be broken to smaller 56 * pages. For 64-bit, kernel text and kernel identity mappings are different, 57 * so we can enable protection checks as well as retain 2MB large page 58 * mappings for kernel text. 59 */ 60#define X86_ALIGN_RODATA_BEGIN . = ALIGN(HPAGE_SIZE); 61 62#define X86_ALIGN_RODATA_END \ 63 . = ALIGN(HPAGE_SIZE); \ 64 __end_rodata_hpage_align = .; \ 65 __end_rodata_aligned = .; 66 67#define ALIGN_ENTRY_TEXT_BEGIN . = ALIGN(PMD_SIZE); 68#define ALIGN_ENTRY_TEXT_END . = ALIGN(PMD_SIZE); 69 70/* 71 * This section contains data which will be mapped as decrypted. Memory 72 * encryption operates on a page basis. Make this section PMD-aligned 73 * to avoid splitting the pages while mapping the section early. 74 * 75 * Note: We use a separate section so that only this section gets 76 * decrypted to avoid exposing more than we wish. 77 */ 78#define BSS_DECRYPTED \ 79 . = ALIGN(PMD_SIZE); \ 80 __start_bss_decrypted = .; \ 81 *(.bss..decrypted); \ 82 . = ALIGN(PAGE_SIZE); \ 83 __start_bss_decrypted_unused = .; \ 84 . = ALIGN(PMD_SIZE); \ 85 __end_bss_decrypted = .; \ 86 87#else 88 89#define X86_ALIGN_RODATA_BEGIN 90#define X86_ALIGN_RODATA_END \ 91 . = ALIGN(PAGE_SIZE); \ 92 __end_rodata_aligned = .; 93 94#define ALIGN_ENTRY_TEXT_BEGIN 95#define ALIGN_ENTRY_TEXT_END 96#define BSS_DECRYPTED 97 98#endif 99#if defined(CONFIG_X86_64) && defined(CONFIG_KEXEC_CORE) 100#define KEXEC_RELOCATE_KERNEL \ 101 . = ALIGN(0x100); \ 102 __relocate_kernel_start = .; \ 103 *(.text..relocate_kernel); \ 104 *(.data..relocate_kernel); \ 105 __relocate_kernel_end = .; 106 107ASSERT(__relocate_kernel_end - __relocate_kernel_start <= KEXEC_CONTROL_CODE_MAX_SIZE, 108 "relocate_kernel code too large!") 109#else 110#define KEXEC_RELOCATE_KERNEL 111#endif 112PHDRS { 113 text PT_LOAD FLAGS(5); /* R_E */ 114 data PT_LOAD FLAGS(6); /* RW_ */ 115#ifdef CONFIG_X86_64 116#ifdef CONFIG_SMP 117 percpu PT_LOAD FLAGS(6); /* RW_ */ 118#endif 119 init PT_LOAD FLAGS(7); /* RWE */ 120#endif 121 note PT_NOTE FLAGS(0); /* ___ */ 122} 123 124SECTIONS 125{ 126 . = __START_KERNEL; 127#ifdef CONFIG_X86_32 128 phys_startup_32 = ABSOLUTE(startup_32 - LOAD_OFFSET); 129#else 130 phys_startup_64 = ABSOLUTE(startup_64 - LOAD_OFFSET); 131#endif 132 133 /* Text and read-only data */ 134 .text : AT(ADDR(.text) - LOAD_OFFSET) { 135 _text = .; 136 _stext = .; 137 ALIGN_ENTRY_TEXT_BEGIN 138 *(.text..__x86.rethunk_untrain) 139 ENTRY_TEXT 140 141#ifdef CONFIG_MITIGATION_SRSO 142 /* 143 * See the comment above srso_alias_untrain_ret()'s 144 * definition. 145 */ 146 . = srso_alias_untrain_ret | (1 << 2) | (1 << 8) | (1 << 14) | (1 << 20); 147 *(.text..__x86.rethunk_safe) 148#endif 149 ALIGN_ENTRY_TEXT_END 150 151 TEXT_TEXT 152 SCHED_TEXT 153 LOCK_TEXT 154 KPROBES_TEXT 155 SOFTIRQENTRY_TEXT 156#ifdef CONFIG_MITIGATION_RETPOLINE 157 *(.text..__x86.indirect_thunk) 158 *(.text..__x86.return_thunk) 159#endif 160 STATIC_CALL_TEXT 161 *(.gnu.warning) 162 163 } :text = 0xcccccccc 164 165 /* bootstrapping code */ 166 .head.text : AT(ADDR(.head.text) - LOAD_OFFSET) { 167 HEAD_TEXT 168 } :text = 0xcccccccc 169 170 /* End of text section, which should occupy whole number of pages */ 171 _etext = .; 172 . = ALIGN(PAGE_SIZE); 173 174 X86_ALIGN_RODATA_BEGIN 175 RO_DATA(PAGE_SIZE) 176 X86_ALIGN_RODATA_END 177 178 /* Data */ 179 .data : AT(ADDR(.data) - LOAD_OFFSET) { 180 /* Start of data section */ 181 _sdata = .; 182 183 /* init_task */ 184 INIT_TASK_DATA(THREAD_SIZE) 185 186 /* equivalent to task_pt_regs(&init_task) */ 187 __top_init_kernel_stack = __end_init_stack - TOP_OF_KERNEL_STACK_PADDING - PTREGS_SIZE; 188 189#ifdef CONFIG_X86_32 190 /* 32 bit has nosave before _edata */ 191 NOSAVE_DATA 192#endif 193 194 PAGE_ALIGNED_DATA(PAGE_SIZE) 195 196 CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES) 197 198 DATA_DATA 199 CONSTRUCTORS 200 KEXEC_RELOCATE_KERNEL 201 202 /* rarely changed data like cpu maps */ 203 READ_MOSTLY_DATA(INTERNODE_CACHE_BYTES) 204 205 /* End of data section */ 206 _edata = .; 207 } :data 208 209 BUG_TABLE 210 211 ORC_UNWIND_TABLE 212 213 /* Init code and data - will be freed after init */ 214 . = ALIGN(PAGE_SIZE); 215 .init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) { 216 __init_begin = .; /* paired with __init_end */ 217 } 218 219#if defined(CONFIG_X86_64) && defined(CONFIG_SMP) 220 /* 221 * percpu offsets are zero-based on SMP. PERCPU_VADDR() changes the 222 * output PHDR, so the next output section - .init.text - should 223 * start another segment - init. 224 */ 225 PERCPU_VADDR(INTERNODE_CACHE_BYTES, 0, :percpu) 226 ASSERT(SIZEOF(.data..percpu) < CONFIG_PHYSICAL_START, 227 "per-CPU data too large - increase CONFIG_PHYSICAL_START") 228#endif 229 230 INIT_TEXT_SECTION(PAGE_SIZE) 231#ifdef CONFIG_X86_64 232 :init 233#endif 234 235 /* 236 * Section for code used exclusively before alternatives are run. All 237 * references to such code must be patched out by alternatives, normally 238 * by using X86_FEATURE_ALWAYS CPU feature bit. 239 * 240 * See static_cpu_has() for an example. 241 */ 242 .altinstr_aux : AT(ADDR(.altinstr_aux) - LOAD_OFFSET) { 243 *(.altinstr_aux) 244 } 245 246 INIT_DATA_SECTION(16) 247 248 .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) { 249 __x86_cpu_dev_start = .; 250 *(.x86_cpu_dev.init) 251 __x86_cpu_dev_end = .; 252 } 253 254#ifdef CONFIG_X86_INTEL_MID 255 .x86_intel_mid_dev.init : AT(ADDR(.x86_intel_mid_dev.init) - \ 256 LOAD_OFFSET) { 257 __x86_intel_mid_dev_start = .; 258 *(.x86_intel_mid_dev.init) 259 __x86_intel_mid_dev_end = .; 260 } 261#endif 262 263#ifdef CONFIG_MITIGATION_RETPOLINE 264 /* 265 * List of instructions that call/jmp/jcc to retpoline thunks 266 * __x86_indirect_thunk_*(). These instructions can be patched along 267 * with alternatives, after which the section can be freed. 268 */ 269 . = ALIGN(8); 270 .retpoline_sites : AT(ADDR(.retpoline_sites) - LOAD_OFFSET) { 271 __retpoline_sites = .; 272 *(.retpoline_sites) 273 __retpoline_sites_end = .; 274 } 275 276 . = ALIGN(8); 277 .return_sites : AT(ADDR(.return_sites) - LOAD_OFFSET) { 278 __return_sites = .; 279 *(.return_sites) 280 __return_sites_end = .; 281 } 282 283 . = ALIGN(8); 284 .call_sites : AT(ADDR(.call_sites) - LOAD_OFFSET) { 285 __call_sites = .; 286 *(.call_sites) 287 __call_sites_end = .; 288 } 289#endif 290 291#ifdef CONFIG_X86_KERNEL_IBT 292 . = ALIGN(8); 293 .ibt_endbr_seal : AT(ADDR(.ibt_endbr_seal) - LOAD_OFFSET) { 294 __ibt_endbr_seal = .; 295 *(.ibt_endbr_seal) 296 __ibt_endbr_seal_end = .; 297 } 298#endif 299 300#ifdef CONFIG_FINEIBT 301 . = ALIGN(8); 302 .cfi_sites : AT(ADDR(.cfi_sites) - LOAD_OFFSET) { 303 __cfi_sites = .; 304 *(.cfi_sites) 305 __cfi_sites_end = .; 306 } 307#endif 308 309 /* 310 * struct alt_inst entries. From the header (alternative.h): 311 * "Alternative instructions for different CPU types or capabilities" 312 * Think locking instructions on spinlocks. 313 */ 314 . = ALIGN(8); 315 .altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) { 316 __alt_instructions = .; 317 *(.altinstructions) 318 __alt_instructions_end = .; 319 } 320 321 /* 322 * And here are the replacement instructions. The linker sticks 323 * them as binary blobs. The .altinstructions has enough data to 324 * get the address and the length of them to patch the kernel safely. 325 */ 326 .altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) { 327 *(.altinstr_replacement) 328 } 329 330 . = ALIGN(8); 331 .apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) { 332 __apicdrivers = .; 333 *(.apicdrivers); 334 __apicdrivers_end = .; 335 } 336 337 . = ALIGN(8); 338 /* 339 * .exit.text is discarded at runtime, not link time, to deal with 340 * references from .altinstructions 341 */ 342 .exit.text : AT(ADDR(.exit.text) - LOAD_OFFSET) { 343 EXIT_TEXT 344 } 345 346 .exit.data : AT(ADDR(.exit.data) - LOAD_OFFSET) { 347 EXIT_DATA 348 } 349 350#if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP) 351 PERCPU_SECTION(INTERNODE_CACHE_BYTES) 352#endif 353 354 RUNTIME_CONST_VARIABLES 355 RUNTIME_CONST(ptr, USER_PTR_MAX) 356 357 . = ALIGN(PAGE_SIZE); 358 359 /* freed after init ends here */ 360 .init.end : AT(ADDR(.init.end) - LOAD_OFFSET) { 361 __init_end = .; 362 } 363 364 /* 365 * smp_locks might be freed after init 366 * start/end must be page aligned 367 */ 368 . = ALIGN(PAGE_SIZE); 369 .smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) { 370 __smp_locks = .; 371 *(.smp_locks) 372 . = ALIGN(PAGE_SIZE); 373 __smp_locks_end = .; 374 } 375 376#ifdef CONFIG_X86_64 377 .data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) { 378 NOSAVE_DATA 379 } 380#endif 381 382 /* BSS */ 383 . = ALIGN(PAGE_SIZE); 384 .bss : AT(ADDR(.bss) - LOAD_OFFSET) { 385 __bss_start = .; 386 *(.bss..page_aligned) 387 . = ALIGN(PAGE_SIZE); 388 *(BSS_MAIN) 389 BSS_DECRYPTED 390 . = ALIGN(PAGE_SIZE); 391 __bss_stop = .; 392 } 393 394 /* 395 * The memory occupied from _text to here, __end_of_kernel_reserve, is 396 * automatically reserved in setup_arch(). Anything after here must be 397 * explicitly reserved using memblock_reserve() or it will be discarded 398 * and treated as available memory. 399 */ 400 __end_of_kernel_reserve = .; 401 402 . = ALIGN(PAGE_SIZE); 403 .brk : AT(ADDR(.brk) - LOAD_OFFSET) { 404 __brk_base = .; 405 . += 64 * 1024; /* 64k alignment slop space */ 406 *(.bss..brk) /* areas brk users have reserved */ 407 __brk_limit = .; 408 } 409 410 . = ALIGN(PAGE_SIZE); /* keep VO_INIT_SIZE page aligned */ 411 _end = .; 412 413#ifdef CONFIG_AMD_MEM_ENCRYPT 414 /* 415 * Early scratch/workarea section: Lives outside of the kernel proper 416 * (_text - _end). 417 * 418 * Resides after _end because even though the .brk section is after 419 * __end_of_kernel_reserve, the .brk section is later reserved as a 420 * part of the kernel. Since it is located after __end_of_kernel_reserve 421 * it will be discarded and become part of the available memory. As 422 * such, it can only be used by very early boot code and must not be 423 * needed afterwards. 424 * 425 * Currently used by SME for performing in-place encryption of the 426 * kernel during boot. Resides on a 2MB boundary to simplify the 427 * pagetable setup used for SME in-place encryption. 428 */ 429 . = ALIGN(HPAGE_SIZE); 430 .init.scratch : AT(ADDR(.init.scratch) - LOAD_OFFSET) { 431 __init_scratch_begin = .; 432 *(.init.scratch) 433 . = ALIGN(HPAGE_SIZE); 434 __init_scratch_end = .; 435 } 436#endif 437 438 STABS_DEBUG 439 DWARF_DEBUG 440#ifdef CONFIG_PROPELLER_CLANG 441 .llvm_bb_addr_map : { *(.llvm_bb_addr_map) } 442#endif 443 444 ELF_DETAILS 445 446 DISCARDS 447 448 /* 449 * Make sure that the .got.plt is either completely empty or it 450 * contains only the lazy dispatch entries. 451 */ 452 .got.plt (INFO) : { *(.got.plt) } 453 ASSERT(SIZEOF(.got.plt) == 0 || 454#ifdef CONFIG_X86_64 455 SIZEOF(.got.plt) == 0x18, 456#else 457 SIZEOF(.got.plt) == 0xc, 458#endif 459 "Unexpected GOT/PLT entries detected!") 460 461 /* 462 * Sections that should stay zero sized, which is safer to 463 * explicitly check instead of blindly discarding. 464 */ 465 .got : { 466 *(.got) *(.igot.*) 467 } 468 ASSERT(SIZEOF(.got) == 0, "Unexpected GOT entries detected!") 469 470 .plt : { 471 *(.plt) *(.plt.*) *(.iplt) 472 } 473 ASSERT(SIZEOF(.plt) == 0, "Unexpected run-time procedure linkages detected!") 474 475 .rel.dyn : { 476 *(.rel.*) *(.rel_*) 477 } 478 ASSERT(SIZEOF(.rel.dyn) == 0, "Unexpected run-time relocations (.rel) detected!") 479 480 .rela.dyn : { 481 *(.rela.*) *(.rela_*) 482 } 483 ASSERT(SIZEOF(.rela.dyn) == 0, "Unexpected run-time relocations (.rela) detected!") 484} 485 486/* 487 * The ASSERT() sink to . is intentional, for binutils 2.14 compatibility: 488 */ 489. = ASSERT((_end - LOAD_OFFSET <= KERNEL_IMAGE_SIZE), 490 "kernel image bigger than KERNEL_IMAGE_SIZE"); 491 492/* needed for Clang - see arch/x86/entry/entry.S */ 493PROVIDE(__ref_stack_chk_guard = __stack_chk_guard); 494 495#ifdef CONFIG_X86_64 496/* 497 * Per-cpu symbols which need to be offset from __per_cpu_load 498 * for the boot processor. 499 */ 500#define INIT_PER_CPU(x) init_per_cpu__##x = ABSOLUTE(x) + __per_cpu_load 501INIT_PER_CPU(gdt_page); 502INIT_PER_CPU(fixed_percpu_data); 503INIT_PER_CPU(irq_stack_backing_store); 504 505#ifdef CONFIG_SMP 506. = ASSERT((fixed_percpu_data == 0), 507 "fixed_percpu_data is not at start of per-cpu area"); 508#endif 509 510#ifdef CONFIG_MITIGATION_UNRET_ENTRY 511. = ASSERT((retbleed_return_thunk & 0x3f) == 0, "retbleed_return_thunk not cacheline-aligned"); 512#endif 513 514#ifdef CONFIG_MITIGATION_SRSO 515. = ASSERT((srso_safe_ret & 0x3f) == 0, "srso_safe_ret not cacheline-aligned"); 516/* 517 * GNU ld cannot do XOR until 2.41. 518 * https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=f6f78318fca803c4907fb8d7f6ded8295f1947b1 519 * 520 * LLVM lld cannot do XOR until lld-17. 521 * https://github.com/llvm/llvm-project/commit/fae96104d4378166cbe5c875ef8ed808a356f3fb 522 * 523 * Instead do: (A | B) - (A & B) in order to compute the XOR 524 * of the two function addresses: 525 */ 526. = ASSERT(((ABSOLUTE(srso_alias_untrain_ret) | srso_alias_safe_ret) - 527 (ABSOLUTE(srso_alias_untrain_ret) & srso_alias_safe_ret)) == ((1 << 2) | (1 << 8) | (1 << 14) | (1 << 20)), 528 "SRSO function pair won't alias"); 529#endif 530 531#endif /* CONFIG_X86_64 */ 532 533/* 534 * The symbols below are referenced using relative relocations in the 535 * respective ELF notes. This produces build time constants that the 536 * linker will never mark as relocatable. (Using just ABSOLUTE() is not 537 * sufficient for that). 538 */ 539#ifdef CONFIG_XEN_PV 540xen_elfnote_entry_value = 541 ABSOLUTE(xen_elfnote_entry) + ABSOLUTE(startup_xen); 542#endif 543#ifdef CONFIG_PVH 544xen_elfnote_phys32_entry_value = 545 ABSOLUTE(xen_elfnote_phys32_entry) + ABSOLUTE(pvh_start_xen - LOAD_OFFSET); 546#endif 547