xref: /linux/arch/x86/kernel/vmlinux.lds.S (revision c5c30a37369313d1f8b84e96e6a4397b4e2b4eb8)
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	__pi___start_bss_decrypted = .;				\
83	*(.bss..decrypted);					\
84	. = ALIGN(PAGE_SIZE);					\
85	__start_bss_decrypted_unused = .;			\
86	. = ALIGN(PMD_SIZE);					\
87	__end_bss_decrypted = .;				\
88	__pi___end_bss_decrypted = .;				\
89
90#else
91
92#define X86_ALIGN_RODATA_BEGIN
93#define X86_ALIGN_RODATA_END					\
94		. = ALIGN(PAGE_SIZE);				\
95		__end_rodata_aligned = .;
96
97#define ALIGN_ENTRY_TEXT_BEGIN
98#define ALIGN_ENTRY_TEXT_END
99#define BSS_DECRYPTED
100
101#endif
102#if defined(CONFIG_X86_64) && defined(CONFIG_KEXEC_CORE)
103#define KEXEC_RELOCATE_KERNEL					\
104	. = ALIGN(0x100);					\
105	__relocate_kernel_start = .;				\
106	*(.text..relocate_kernel);				\
107	*(.data..relocate_kernel);				\
108	__relocate_kernel_end = .;
109
110ASSERT(__relocate_kernel_end - __relocate_kernel_start <= KEXEC_CONTROL_CODE_MAX_SIZE,
111	"relocate_kernel code too large!")
112#else
113#define KEXEC_RELOCATE_KERNEL
114#endif
115PHDRS {
116	text PT_LOAD FLAGS(5);          /* R_E */
117	data PT_LOAD FLAGS(6);          /* RW_ */
118	note PT_NOTE FLAGS(0);          /* ___ */
119}
120
121SECTIONS
122{
123	. = __START_KERNEL;
124#ifdef CONFIG_X86_32
125	phys_startup_32 = ABSOLUTE(startup_32 - LOAD_OFFSET);
126#else
127	phys_startup_64 = ABSOLUTE(startup_64 - LOAD_OFFSET);
128#endif
129
130	/* Text and read-only data */
131	.text :  AT(ADDR(.text) - LOAD_OFFSET) {
132		_text = .;
133		__pi__text = .;
134		_stext = .;
135		ALIGN_ENTRY_TEXT_BEGIN
136		*(.text..__x86.rethunk_untrain)
137		ENTRY_TEXT
138
139#ifdef CONFIG_MITIGATION_SRSO
140		/*
141		 * See the comment above srso_alias_untrain_ret()'s
142		 * definition.
143		 */
144		. = srso_alias_untrain_ret | (1 << 2) | (1 << 8) | (1 << 14) | (1 << 20);
145		*(.text..__x86.rethunk_safe)
146#endif
147		ALIGN_ENTRY_TEXT_END
148
149		TEXT_TEXT
150		SCHED_TEXT
151		LOCK_TEXT
152		KPROBES_TEXT
153		SOFTIRQENTRY_TEXT
154#ifdef CONFIG_MITIGATION_RETPOLINE
155		*(.text..__x86.indirect_thunk)
156		*(.text..__x86.return_thunk)
157#endif
158		STATIC_CALL_TEXT
159		*(.gnu.warning)
160
161	} :text = 0xcccccccc
162
163	/* bootstrapping code */
164	.head.text : AT(ADDR(.head.text) - LOAD_OFFSET) {
165		HEAD_TEXT
166	} :text = 0xcccccccc
167
168	/* End of text section, which should occupy whole number of pages */
169	_etext = .;
170	. = ALIGN(PAGE_SIZE);
171
172	X86_ALIGN_RODATA_BEGIN
173	RO_DATA(PAGE_SIZE)
174	X86_ALIGN_RODATA_END
175
176	/* Data */
177	.data : AT(ADDR(.data) - LOAD_OFFSET) {
178		/* Start of data section */
179		_sdata = .;
180
181		/* init_task */
182		INIT_TASK_DATA(THREAD_SIZE)
183
184		/* equivalent to task_pt_regs(&init_task) */
185		__top_init_kernel_stack = __end_init_stack - TOP_OF_KERNEL_STACK_PADDING - PTREGS_SIZE;
186
187#ifdef CONFIG_X86_32
188		/* 32 bit has nosave before _edata */
189		NOSAVE_DATA
190#endif
191
192		PAGE_ALIGNED_DATA(PAGE_SIZE)
193
194		CACHE_HOT_DATA(L1_CACHE_BYTES)
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	INIT_TEXT_SECTION(PAGE_SIZE)
220
221	/*
222	 * Section for code used exclusively before alternatives are run. All
223	 * references to such code must be patched out by alternatives, normally
224	 * by using X86_FEATURE_ALWAYS CPU feature bit.
225	 *
226	 * See static_cpu_has() for an example.
227	 */
228	.altinstr_aux : AT(ADDR(.altinstr_aux) - LOAD_OFFSET) {
229		*(.altinstr_aux)
230		. = ALIGN(PAGE_SIZE);
231		__inittext_end = .;
232	}
233
234	INIT_DATA_SECTION(16)
235
236	.x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
237		__x86_cpu_dev_start = .;
238		*(.x86_cpu_dev.init)
239		__x86_cpu_dev_end = .;
240	}
241
242#ifdef CONFIG_X86_INTEL_MID
243	.x86_intel_mid_dev.init : AT(ADDR(.x86_intel_mid_dev.init) - \
244								LOAD_OFFSET) {
245		__x86_intel_mid_dev_start = .;
246		*(.x86_intel_mid_dev.init)
247		__x86_intel_mid_dev_end = .;
248	}
249#endif
250
251#ifdef CONFIG_MITIGATION_RETPOLINE
252	/*
253	 * List of instructions that call/jmp/jcc to retpoline thunks
254	 * __x86_indirect_thunk_*(). These instructions can be patched along
255	 * with alternatives, after which the section can be freed.
256	 */
257	. = ALIGN(8);
258	.retpoline_sites : AT(ADDR(.retpoline_sites) - LOAD_OFFSET) {
259		__retpoline_sites = .;
260		*(.retpoline_sites)
261		__retpoline_sites_end = .;
262	}
263
264	. = ALIGN(8);
265	.return_sites : AT(ADDR(.return_sites) - LOAD_OFFSET) {
266		__return_sites = .;
267		*(.return_sites)
268		__return_sites_end = .;
269	}
270
271	. = ALIGN(8);
272	.call_sites : AT(ADDR(.call_sites) - LOAD_OFFSET) {
273		__call_sites = .;
274		*(.call_sites)
275		__call_sites_end = .;
276	}
277#endif
278
279#ifdef CONFIG_X86_KERNEL_IBT
280	. = ALIGN(8);
281	.ibt_endbr_seal : AT(ADDR(.ibt_endbr_seal) - LOAD_OFFSET) {
282		__ibt_endbr_seal = .;
283		*(.ibt_endbr_seal)
284		__ibt_endbr_seal_end = .;
285	}
286#endif
287
288#ifdef CONFIG_FINEIBT
289	. = ALIGN(8);
290	.cfi_sites : AT(ADDR(.cfi_sites) - LOAD_OFFSET) {
291		__cfi_sites = .;
292		*(.cfi_sites)
293		__cfi_sites_end = .;
294	}
295#endif
296
297	/*
298	 * struct alt_inst entries. From the header (alternative.h):
299	 * "Alternative instructions for different CPU types or capabilities"
300	 * Think locking instructions on spinlocks.
301	 */
302	. = ALIGN(8);
303	.altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) {
304		__alt_instructions = .;
305		*(.altinstructions)
306		__alt_instructions_end = .;
307	}
308
309	/*
310	 * And here are the replacement instructions. The linker sticks
311	 * them as binary blobs. The .altinstructions has enough data to
312	 * get the address and the length of them to patch the kernel safely.
313	 */
314	.altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) {
315		*(.altinstr_replacement)
316	}
317
318	. = ALIGN(8);
319	.apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) {
320		__apicdrivers = .;
321		*(.apicdrivers);
322		__apicdrivers_end = .;
323	}
324
325	. = ALIGN(8);
326	/*
327	 * .exit.text is discarded at runtime, not link time, to deal with
328	 *  references from .altinstructions
329	 */
330	.exit.text : AT(ADDR(.exit.text) - LOAD_OFFSET) {
331		EXIT_TEXT
332	}
333
334	.exit.data : AT(ADDR(.exit.data) - LOAD_OFFSET) {
335		EXIT_DATA
336	}
337
338	PERCPU_SECTION(L1_CACHE_BYTES)
339	ASSERT(__per_cpu_hot_end - __per_cpu_hot_start <= 64, "percpu cache hot data too large")
340
341	RUNTIME_CONST_VARIABLES
342	RUNTIME_CONST(ptr, USER_PTR_MAX)
343
344	. = ALIGN(PAGE_SIZE);
345
346	/* freed after init ends here */
347	.init.end : AT(ADDR(.init.end) - LOAD_OFFSET) {
348		__init_end = .;
349	}
350
351	/*
352	 * smp_locks might be freed after init
353	 * start/end must be page aligned
354	 */
355	. = ALIGN(PAGE_SIZE);
356	.smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) {
357		__smp_locks = .;
358		*(.smp_locks)
359		. = ALIGN(PAGE_SIZE);
360		__smp_locks_end = .;
361	}
362
363#ifdef CONFIG_X86_64
364	.data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) {
365		NOSAVE_DATA
366	}
367#endif
368
369	/* BSS */
370	. = ALIGN(PAGE_SIZE);
371	.bss : AT(ADDR(.bss) - LOAD_OFFSET) {
372		__bss_start = .;
373		*(.bss..page_aligned)
374		. = ALIGN(PAGE_SIZE);
375		*(BSS_MAIN)
376		BSS_DECRYPTED
377		. = ALIGN(PAGE_SIZE);
378		__bss_stop = .;
379	}
380
381	/*
382	 * The memory occupied from _text to here, __end_of_kernel_reserve, is
383	 * automatically reserved in setup_arch(). Anything after here must be
384	 * explicitly reserved using memblock_reserve() or it will be discarded
385	 * and treated as available memory.
386	 */
387	__end_of_kernel_reserve = .;
388
389	. = ALIGN(PAGE_SIZE);
390	.brk : AT(ADDR(.brk) - LOAD_OFFSET) {
391		__brk_base = .;
392		. += 64 * 1024;		/* 64k alignment slop space */
393		*(.bss..brk)		/* areas brk users have reserved */
394		__brk_limit = .;
395	}
396
397	. = ALIGN(PAGE_SIZE);		/* keep VO_INIT_SIZE page aligned */
398	_end = .;
399	__pi__end = .;
400
401#ifdef CONFIG_AMD_MEM_ENCRYPT
402	/*
403	 * Early scratch/workarea section: Lives outside of the kernel proper
404	 * (_text - _end).
405	 *
406	 * Resides after _end because even though the .brk section is after
407	 * __end_of_kernel_reserve, the .brk section is later reserved as a
408	 * part of the kernel. Since it is located after __end_of_kernel_reserve
409	 * it will be discarded and become part of the available memory. As
410	 * such, it can only be used by very early boot code and must not be
411	 * needed afterwards.
412	 *
413	 * Currently used by SME for performing in-place encryption of the
414	 * kernel during boot. Resides on a 2MB boundary to simplify the
415	 * pagetable setup used for SME in-place encryption.
416	 */
417	. = ALIGN(HPAGE_SIZE);
418	.init.scratch : AT(ADDR(.init.scratch) - LOAD_OFFSET) {
419		__init_scratch_begin = .;
420		*(.init.scratch)
421		. = ALIGN(HPAGE_SIZE);
422		__init_scratch_end = .;
423	}
424#endif
425
426	STABS_DEBUG
427	DWARF_DEBUG
428#ifdef CONFIG_PROPELLER_CLANG
429	.llvm_bb_addr_map : { *(.llvm_bb_addr_map) }
430#endif
431
432	ELF_DETAILS
433
434	DISCARDS
435
436	/*
437	 * Make sure that the .got.plt is either completely empty or it
438	 * contains only the lazy dispatch entries.
439	 */
440	.got.plt (INFO) : { *(.got.plt) }
441	ASSERT(SIZEOF(.got.plt) == 0 ||
442#ifdef CONFIG_X86_64
443	       SIZEOF(.got.plt) == 0x18,
444#else
445	       SIZEOF(.got.plt) == 0xc,
446#endif
447	       "Unexpected GOT/PLT entries detected!")
448
449	/*
450	 * Sections that should stay zero sized, which is safer to
451	 * explicitly check instead of blindly discarding.
452	 */
453	.got : {
454		*(.got) *(.igot.*)
455	}
456	ASSERT(SIZEOF(.got) == 0, "Unexpected GOT entries detected!")
457
458	.plt : {
459		*(.plt) *(.plt.*) *(.iplt)
460	}
461	ASSERT(SIZEOF(.plt) == 0, "Unexpected run-time procedure linkages detected!")
462
463	.rel.dyn : {
464		*(.rel.*) *(.rel_*)
465	}
466	ASSERT(SIZEOF(.rel.dyn) == 0, "Unexpected run-time relocations (.rel) detected!")
467
468	.rela.dyn : {
469		*(.rela.*) *(.rela_*)
470	}
471	ASSERT(SIZEOF(.rela.dyn) == 0, "Unexpected run-time relocations (.rela) detected!")
472}
473
474/*
475 * COMPILE_TEST kernels can be large - CONFIG_KASAN, for example, can cause
476 * this.  Let's assume that nobody will be running a COMPILE_TEST kernel and
477 * let's assert that fuller build coverage is more valuable than being able to
478 * run a COMPILE_TEST kernel.
479 */
480#ifndef CONFIG_COMPILE_TEST
481/*
482 * The ASSERT() sync to . is intentional, for binutils 2.14 compatibility:
483 */
484. = ASSERT((_end - LOAD_OFFSET <= KERNEL_IMAGE_SIZE),
485	   "kernel image bigger than KERNEL_IMAGE_SIZE");
486#endif
487
488/* needed for Clang - see arch/x86/entry/entry.S */
489PROVIDE(__ref_stack_chk_guard = __stack_chk_guard);
490
491#ifdef CONFIG_X86_64
492
493#ifdef CONFIG_MITIGATION_UNRET_ENTRY
494. = ASSERT((retbleed_return_thunk & 0x3f) == 0, "retbleed_return_thunk not cacheline-aligned");
495#endif
496
497#ifdef CONFIG_MITIGATION_SRSO
498. = ASSERT((srso_safe_ret & 0x3f) == 0, "srso_safe_ret not cacheline-aligned");
499/*
500 * GNU ld cannot do XOR until 2.41.
501 * https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=f6f78318fca803c4907fb8d7f6ded8295f1947b1
502 *
503 * LLVM lld cannot do XOR until lld-17.
504 * https://github.com/llvm/llvm-project/commit/fae96104d4378166cbe5c875ef8ed808a356f3fb
505 *
506 * Instead do: (A | B) - (A & B) in order to compute the XOR
507 * of the two function addresses:
508 */
509. = ASSERT(((ABSOLUTE(srso_alias_untrain_ret) | srso_alias_safe_ret) -
510		(ABSOLUTE(srso_alias_untrain_ret) & srso_alias_safe_ret)) == ((1 << 2) | (1 << 8) | (1 << 14) | (1 << 20)),
511		"SRSO function pair won't alias");
512#endif
513
514#if defined(CONFIG_MITIGATION_ITS) && !defined(CONFIG_DEBUG_FORCE_FUNCTION_ALIGN_64B)
515. = ASSERT(__x86_indirect_its_thunk_rax & 0x20, "__x86_indirect_thunk_rax not in second half of cacheline");
516. = ASSERT(((__x86_indirect_its_thunk_rcx - __x86_indirect_its_thunk_rax) % 64) == 0, "Indirect thunks are not cacheline apart");
517. = ASSERT(__x86_indirect_its_thunk_array == __x86_indirect_its_thunk_rax, "Gap in ITS thunk array");
518#endif
519
520#if defined(CONFIG_MITIGATION_ITS) && !defined(CONFIG_DEBUG_FORCE_FUNCTION_ALIGN_64B)
521. = ASSERT(its_return_thunk & 0x20, "its_return_thunk not in second half of cacheline");
522#endif
523
524#endif /* CONFIG_X86_64 */
525
526/*
527 * The symbols below are referenced using relative relocations in the
528 * respective ELF notes. This produces build time constants that the
529 * linker will never mark as relocatable. (Using just ABSOLUTE() is not
530 * sufficient for that).
531 */
532#ifdef CONFIG_XEN_PV
533xen_elfnote_entry_value =
534	ABSOLUTE(xen_elfnote_entry) + ABSOLUTE(startup_xen);
535#endif
536#ifdef CONFIG_PVH
537xen_elfnote_phys32_entry_value =
538	ABSOLUTE(xen_elfnote_phys32_entry) + ABSOLUTE(pvh_start_xen - LOAD_OFFSET);
539#endif
540
541#include "../boot/startup/exports.h"
542