xref: /openbmc/linux/arch/x86/kernel/vmlinux.lds.S (revision 7288dd2f)
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#ifdef CONFIG_X86_32
19#define LOAD_OFFSET __PAGE_OFFSET
20#else
21#define LOAD_OFFSET __START_KERNEL_map
22#endif
23
24#define RUNTIME_DISCARD_EXIT
25#define EMITS_PT_NOTE
26#define RO_EXCEPTION_TABLE_ALIGN	16
27
28#include <asm-generic/vmlinux.lds.h>
29#include <asm/asm-offsets.h>
30#include <asm/thread_info.h>
31#include <asm/page_types.h>
32#include <asm/orc_lookup.h>
33#include <asm/cache.h>
34#include <asm/boot.h>
35
36#undef i386     /* in case the preprocessor is a 32bit one */
37
38OUTPUT_FORMAT(CONFIG_OUTPUT_FORMAT)
39
40#ifdef CONFIG_X86_32
41OUTPUT_ARCH(i386)
42ENTRY(phys_startup_32)
43#else
44OUTPUT_ARCH(i386:x86-64)
45ENTRY(phys_startup_64)
46#endif
47
48jiffies = jiffies_64;
49
50#if defined(CONFIG_X86_64)
51/*
52 * On 64-bit, align RODATA to 2MB so we retain large page mappings for
53 * boundaries spanning kernel text, rodata and data sections.
54 *
55 * However, kernel identity mappings will have different RWX permissions
56 * to the pages mapping to text and to the pages padding (which are freed) the
57 * text section. Hence kernel identity mappings will be broken to smaller
58 * pages. For 64-bit, kernel text and kernel identity mappings are different,
59 * so we can enable protection checks as well as retain 2MB large page
60 * mappings for kernel text.
61 */
62#define X86_ALIGN_RODATA_BEGIN	. = ALIGN(HPAGE_SIZE);
63
64#define X86_ALIGN_RODATA_END					\
65		. = ALIGN(HPAGE_SIZE);				\
66		__end_rodata_hpage_align = .;			\
67		__end_rodata_aligned = .;
68
69#define ALIGN_ENTRY_TEXT_BEGIN	. = ALIGN(PMD_SIZE);
70#define ALIGN_ENTRY_TEXT_END	. = ALIGN(PMD_SIZE);
71
72/*
73 * This section contains data which will be mapped as decrypted. Memory
74 * encryption operates on a page basis. Make this section PMD-aligned
75 * to avoid splitting the pages while mapping the section early.
76 *
77 * Note: We use a separate section so that only this section gets
78 * decrypted to avoid exposing more than we wish.
79 */
80#define BSS_DECRYPTED						\
81	. = ALIGN(PMD_SIZE);					\
82	__start_bss_decrypted = .;				\
83	*(.bss..decrypted);					\
84	. = ALIGN(PAGE_SIZE);					\
85	__start_bss_decrypted_unused = .;			\
86	. = ALIGN(PMD_SIZE);					\
87	__end_bss_decrypted = .;				\
88
89#else
90
91#define X86_ALIGN_RODATA_BEGIN
92#define X86_ALIGN_RODATA_END					\
93		. = ALIGN(PAGE_SIZE);				\
94		__end_rodata_aligned = .;
95
96#define ALIGN_ENTRY_TEXT_BEGIN
97#define ALIGN_ENTRY_TEXT_END
98#define BSS_DECRYPTED
99
100#endif
101
102PHDRS {
103	text PT_LOAD FLAGS(5);          /* R_E */
104	data PT_LOAD FLAGS(6);          /* RW_ */
105#ifdef CONFIG_X86_64
106#ifdef CONFIG_SMP
107	percpu PT_LOAD FLAGS(6);        /* RW_ */
108#endif
109	init PT_LOAD FLAGS(7);          /* RWE */
110#endif
111	note PT_NOTE FLAGS(0);          /* ___ */
112}
113
114SECTIONS
115{
116#ifdef CONFIG_X86_32
117	. = LOAD_OFFSET + LOAD_PHYSICAL_ADDR;
118	phys_startup_32 = ABSOLUTE(startup_32 - LOAD_OFFSET);
119#else
120	. = __START_KERNEL;
121	phys_startup_64 = ABSOLUTE(startup_64 - LOAD_OFFSET);
122#endif
123
124	/* Text and read-only data */
125	.text :  AT(ADDR(.text) - LOAD_OFFSET) {
126		_text = .;
127		_stext = .;
128		/* bootstrapping code */
129		HEAD_TEXT
130		TEXT_TEXT
131		SCHED_TEXT
132		LOCK_TEXT
133		KPROBES_TEXT
134		SOFTIRQENTRY_TEXT
135#ifdef CONFIG_RETPOLINE
136		__indirect_thunk_start = .;
137		*(.text.__x86.indirect_thunk)
138		*(.text.__x86.return_thunk)
139		__indirect_thunk_end = .;
140#endif
141		STATIC_CALL_TEXT
142
143		ALIGN_ENTRY_TEXT_BEGIN
144#ifdef CONFIG_CPU_SRSO
145		*(.text.__x86.rethunk_untrain)
146#endif
147
148		ENTRY_TEXT
149
150#ifdef CONFIG_CPU_SRSO
151		/*
152		 * See the comment above srso_untrain_ret_alias()'s
153		 * definition.
154		 */
155		. = srso_untrain_ret_alias | (1 << 2) | (1 << 8) | (1 << 14) | (1 << 20);
156		*(.text.__x86.rethunk_safe)
157#endif
158		ALIGN_ENTRY_TEXT_END
159		*(.gnu.warning)
160
161	} :text =0xcccc
162
163	/* End of text section, which should occupy whole number of pages */
164	_etext = .;
165	. = ALIGN(PAGE_SIZE);
166
167	X86_ALIGN_RODATA_BEGIN
168	RO_DATA(PAGE_SIZE)
169	X86_ALIGN_RODATA_END
170
171	/* Data */
172	.data : AT(ADDR(.data) - LOAD_OFFSET) {
173		/* Start of data section */
174		_sdata = .;
175
176		/* init_task */
177		INIT_TASK_DATA(THREAD_SIZE)
178
179#ifdef CONFIG_X86_32
180		/* 32 bit has nosave before _edata */
181		NOSAVE_DATA
182#endif
183
184		PAGE_ALIGNED_DATA(PAGE_SIZE)
185
186		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
187
188		DATA_DATA
189		CONSTRUCTORS
190
191		/* rarely changed data like cpu maps */
192		READ_MOSTLY_DATA(INTERNODE_CACHE_BYTES)
193
194		/* End of data section */
195		_edata = .;
196	} :data
197
198	BUG_TABLE
199
200	ORC_UNWIND_TABLE
201
202	. = ALIGN(PAGE_SIZE);
203	__vvar_page = .;
204
205	.vvar : AT(ADDR(.vvar) - LOAD_OFFSET) {
206		/* work around gold bug 13023 */
207		__vvar_beginning_hack = .;
208
209		/* Place all vvars at the offsets in asm/vvar.h. */
210#define EMIT_VVAR(name, offset)				\
211		. = __vvar_beginning_hack + offset;	\
212		*(.vvar_ ## name)
213#include <asm/vvar.h>
214#undef EMIT_VVAR
215
216		/*
217		 * Pad the rest of the page with zeros.  Otherwise the loader
218		 * can leave garbage here.
219		 */
220		. = __vvar_beginning_hack + PAGE_SIZE;
221	} :data
222
223	. = ALIGN(__vvar_page + PAGE_SIZE, PAGE_SIZE);
224
225	/* Init code and data - will be freed after init */
226	. = ALIGN(PAGE_SIZE);
227	.init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) {
228		__init_begin = .; /* paired with __init_end */
229	}
230
231#if defined(CONFIG_X86_64) && defined(CONFIG_SMP)
232	/*
233	 * percpu offsets are zero-based on SMP.  PERCPU_VADDR() changes the
234	 * output PHDR, so the next output section - .init.text - should
235	 * start another segment - init.
236	 */
237	PERCPU_VADDR(INTERNODE_CACHE_BYTES, 0, :percpu)
238	ASSERT(SIZEOF(.data..percpu) < CONFIG_PHYSICAL_START,
239	       "per-CPU data too large - increase CONFIG_PHYSICAL_START")
240#endif
241
242	INIT_TEXT_SECTION(PAGE_SIZE)
243#ifdef CONFIG_X86_64
244	:init
245#endif
246
247	/*
248	 * Section for code used exclusively before alternatives are run. All
249	 * references to such code must be patched out by alternatives, normally
250	 * by using X86_FEATURE_ALWAYS CPU feature bit.
251	 *
252	 * See static_cpu_has() for an example.
253	 */
254	.altinstr_aux : AT(ADDR(.altinstr_aux) - LOAD_OFFSET) {
255		*(.altinstr_aux)
256	}
257
258	INIT_DATA_SECTION(16)
259
260	.x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
261		__x86_cpu_dev_start = .;
262		*(.x86_cpu_dev.init)
263		__x86_cpu_dev_end = .;
264	}
265
266#ifdef CONFIG_X86_INTEL_MID
267	.x86_intel_mid_dev.init : AT(ADDR(.x86_intel_mid_dev.init) - \
268								LOAD_OFFSET) {
269		__x86_intel_mid_dev_start = .;
270		*(.x86_intel_mid_dev.init)
271		__x86_intel_mid_dev_end = .;
272	}
273#endif
274
275	/*
276	 * start address and size of operations which during runtime
277	 * can be patched with virtualization friendly instructions or
278	 * baremetal native ones. Think page table operations.
279	 * Details in paravirt_types.h
280	 */
281	. = ALIGN(8);
282	.parainstructions : AT(ADDR(.parainstructions) - LOAD_OFFSET) {
283		__parainstructions = .;
284		*(.parainstructions)
285		__parainstructions_end = .;
286	}
287
288#ifdef CONFIG_RETPOLINE
289	/*
290	 * List of instructions that call/jmp/jcc to retpoline thunks
291	 * __x86_indirect_thunk_*(). These instructions can be patched along
292	 * with alternatives, after which the section can be freed.
293	 */
294	. = ALIGN(8);
295	.retpoline_sites : AT(ADDR(.retpoline_sites) - LOAD_OFFSET) {
296		__retpoline_sites = .;
297		*(.retpoline_sites)
298		__retpoline_sites_end = .;
299	}
300
301	. = ALIGN(8);
302	.return_sites : AT(ADDR(.return_sites) - LOAD_OFFSET) {
303		__return_sites = .;
304		*(.return_sites)
305		__return_sites_end = .;
306	}
307
308	. = ALIGN(8);
309	.call_sites : AT(ADDR(.call_sites) - LOAD_OFFSET) {
310		__call_sites = .;
311		*(.call_sites)
312		__call_sites_end = .;
313	}
314#endif
315
316#ifdef CONFIG_X86_KERNEL_IBT
317	. = ALIGN(8);
318	.ibt_endbr_seal : AT(ADDR(.ibt_endbr_seal) - LOAD_OFFSET) {
319		__ibt_endbr_seal = .;
320		*(.ibt_endbr_seal)
321		__ibt_endbr_seal_end = .;
322	}
323#endif
324
325#ifdef CONFIG_FINEIBT
326	. = ALIGN(8);
327	.cfi_sites : AT(ADDR(.cfi_sites) - LOAD_OFFSET) {
328		__cfi_sites = .;
329		*(.cfi_sites)
330		__cfi_sites_end = .;
331	}
332#endif
333
334	/*
335	 * struct alt_inst entries. From the header (alternative.h):
336	 * "Alternative instructions for different CPU types or capabilities"
337	 * Think locking instructions on spinlocks.
338	 */
339	. = ALIGN(8);
340	.altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) {
341		__alt_instructions = .;
342		*(.altinstructions)
343		__alt_instructions_end = .;
344	}
345
346	/*
347	 * And here are the replacement instructions. The linker sticks
348	 * them as binary blobs. The .altinstructions has enough data to
349	 * get the address and the length of them to patch the kernel safely.
350	 */
351	.altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) {
352		*(.altinstr_replacement)
353	}
354
355	. = ALIGN(8);
356	.apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) {
357		__apicdrivers = .;
358		*(.apicdrivers);
359		__apicdrivers_end = .;
360	}
361
362	. = ALIGN(8);
363	/*
364	 * .exit.text is discarded at runtime, not link time, to deal with
365	 *  references from .altinstructions
366	 */
367	.exit.text : AT(ADDR(.exit.text) - LOAD_OFFSET) {
368		EXIT_TEXT
369	}
370
371	.exit.data : AT(ADDR(.exit.data) - LOAD_OFFSET) {
372		EXIT_DATA
373	}
374
375#if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP)
376	PERCPU_SECTION(INTERNODE_CACHE_BYTES)
377#endif
378
379	. = ALIGN(PAGE_SIZE);
380
381	/* freed after init ends here */
382	.init.end : AT(ADDR(.init.end) - LOAD_OFFSET) {
383		__init_end = .;
384	}
385
386	/*
387	 * smp_locks might be freed after init
388	 * start/end must be page aligned
389	 */
390	. = ALIGN(PAGE_SIZE);
391	.smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) {
392		__smp_locks = .;
393		*(.smp_locks)
394		. = ALIGN(PAGE_SIZE);
395		__smp_locks_end = .;
396	}
397
398#ifdef CONFIG_X86_64
399	.data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) {
400		NOSAVE_DATA
401	}
402#endif
403
404	/* BSS */
405	. = ALIGN(PAGE_SIZE);
406	.bss : AT(ADDR(.bss) - LOAD_OFFSET) {
407		__bss_start = .;
408		*(.bss..page_aligned)
409		. = ALIGN(PAGE_SIZE);
410		*(BSS_MAIN)
411		BSS_DECRYPTED
412		. = ALIGN(PAGE_SIZE);
413		__bss_stop = .;
414	}
415
416	/*
417	 * The memory occupied from _text to here, __end_of_kernel_reserve, is
418	 * automatically reserved in setup_arch(). Anything after here must be
419	 * explicitly reserved using memblock_reserve() or it will be discarded
420	 * and treated as available memory.
421	 */
422	__end_of_kernel_reserve = .;
423
424	. = ALIGN(PAGE_SIZE);
425	.brk : AT(ADDR(.brk) - LOAD_OFFSET) {
426		__brk_base = .;
427		. += 64 * 1024;		/* 64k alignment slop space */
428		*(.bss..brk)		/* areas brk users have reserved */
429		__brk_limit = .;
430	}
431
432	. = ALIGN(PAGE_SIZE);		/* keep VO_INIT_SIZE page aligned */
433	_end = .;
434
435#ifdef CONFIG_AMD_MEM_ENCRYPT
436	/*
437	 * Early scratch/workarea section: Lives outside of the kernel proper
438	 * (_text - _end).
439	 *
440	 * Resides after _end because even though the .brk section is after
441	 * __end_of_kernel_reserve, the .brk section is later reserved as a
442	 * part of the kernel. Since it is located after __end_of_kernel_reserve
443	 * it will be discarded and become part of the available memory. As
444	 * such, it can only be used by very early boot code and must not be
445	 * needed afterwards.
446	 *
447	 * Currently used by SME for performing in-place encryption of the
448	 * kernel during boot. Resides on a 2MB boundary to simplify the
449	 * pagetable setup used for SME in-place encryption.
450	 */
451	. = ALIGN(HPAGE_SIZE);
452	.init.scratch : AT(ADDR(.init.scratch) - LOAD_OFFSET) {
453		__init_scratch_begin = .;
454		*(.init.scratch)
455		. = ALIGN(HPAGE_SIZE);
456		__init_scratch_end = .;
457	}
458#endif
459
460	STABS_DEBUG
461	DWARF_DEBUG
462	ELF_DETAILS
463
464	DISCARDS
465
466	/*
467	 * Make sure that the .got.plt is either completely empty or it
468	 * contains only the lazy dispatch entries.
469	 */
470	.got.plt (INFO) : { *(.got.plt) }
471	ASSERT(SIZEOF(.got.plt) == 0 ||
472#ifdef CONFIG_X86_64
473	       SIZEOF(.got.plt) == 0x18,
474#else
475	       SIZEOF(.got.plt) == 0xc,
476#endif
477	       "Unexpected GOT/PLT entries detected!")
478
479	/*
480	 * Sections that should stay zero sized, which is safer to
481	 * explicitly check instead of blindly discarding.
482	 */
483	.got : {
484		*(.got) *(.igot.*)
485	}
486	ASSERT(SIZEOF(.got) == 0, "Unexpected GOT entries detected!")
487
488	.plt : {
489		*(.plt) *(.plt.*) *(.iplt)
490	}
491	ASSERT(SIZEOF(.plt) == 0, "Unexpected run-time procedure linkages detected!")
492
493	.rel.dyn : {
494		*(.rel.*) *(.rel_*)
495	}
496	ASSERT(SIZEOF(.rel.dyn) == 0, "Unexpected run-time relocations (.rel) detected!")
497
498	.rela.dyn : {
499		*(.rela.*) *(.rela_*)
500	}
501	ASSERT(SIZEOF(.rela.dyn) == 0, "Unexpected run-time relocations (.rela) detected!")
502}
503
504/*
505 * The ASSERT() sink to . is intentional, for binutils 2.14 compatibility:
506 */
507. = ASSERT((_end - LOAD_OFFSET <= KERNEL_IMAGE_SIZE),
508	   "kernel image bigger than KERNEL_IMAGE_SIZE");
509
510#ifdef CONFIG_X86_64
511/*
512 * Per-cpu symbols which need to be offset from __per_cpu_load
513 * for the boot processor.
514 */
515#define INIT_PER_CPU(x) init_per_cpu__##x = ABSOLUTE(x) + __per_cpu_load
516INIT_PER_CPU(gdt_page);
517INIT_PER_CPU(fixed_percpu_data);
518INIT_PER_CPU(irq_stack_backing_store);
519
520#ifdef CONFIG_SMP
521. = ASSERT((fixed_percpu_data == 0),
522           "fixed_percpu_data is not at start of per-cpu area");
523#endif
524
525#ifdef CONFIG_RETHUNK
526. = ASSERT((__ret & 0x3f) == 0, "__ret not cacheline-aligned");
527. = ASSERT((srso_safe_ret & 0x3f) == 0, "srso_safe_ret not cacheline-aligned");
528#endif
529
530#ifdef CONFIG_CPU_SRSO
531/*
532 * GNU ld cannot do XOR until 2.41.
533 * https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=f6f78318fca803c4907fb8d7f6ded8295f1947b1
534 *
535 * LLVM lld cannot do XOR until lld-17.
536 * https://github.com/llvm/llvm-project/commit/fae96104d4378166cbe5c875ef8ed808a356f3fb
537 *
538 * Instead do: (A | B) - (A & B) in order to compute the XOR
539 * of the two function addresses:
540 */
541. = ASSERT(((ABSOLUTE(srso_untrain_ret_alias) | srso_safe_ret_alias) -
542		(ABSOLUTE(srso_untrain_ret_alias) & srso_safe_ret_alias)) == ((1 << 2) | (1 << 8) | (1 << 14) | (1 << 20)),
543		"SRSO function pair won't alias");
544#endif
545
546#endif /* CONFIG_X86_64 */
547