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#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
32#undef i386     /* in case the preprocessor is a 32bit one */
33
34OUTPUT_FORMAT(CONFIG_OUTPUT_FORMAT)
35
36#ifdef CONFIG_X86_32
37OUTPUT_ARCH(i386)
38ENTRY(phys_startup_32)
39jiffies = jiffies_64;
40#else
41OUTPUT_ARCH(i386:x86-64)
42ENTRY(phys_startup_64)
43jiffies_64 = jiffies;
44#endif
45
46#if defined(CONFIG_X86_64)
47/*
48 * On 64-bit, align RODATA to 2MB so we retain large page mappings for
49 * boundaries spanning kernel text, rodata and data sections.
50 *
51 * However, kernel identity mappings will have different RWX permissions
52 * to the pages mapping to text and to the pages padding (which are freed) the
53 * text section. Hence kernel identity mappings will be broken to smaller
54 * pages. For 64-bit, kernel text and kernel identity mappings are different,
55 * so we can enable protection checks as well as retain 2MB large page
56 * mappings for kernel text.
57 */
58#define X86_ALIGN_RODATA_BEGIN	. = ALIGN(HPAGE_SIZE);
59
60#define X86_ALIGN_RODATA_END					\
61		. = ALIGN(HPAGE_SIZE);				\
62		__end_rodata_hpage_align = .;			\
63		__end_rodata_aligned = .;
64
65#define ALIGN_ENTRY_TEXT_BEGIN	. = ALIGN(PMD_SIZE);
66#define ALIGN_ENTRY_TEXT_END	. = ALIGN(PMD_SIZE);
67
68/*
69 * This section contains data which will be mapped as decrypted. Memory
70 * encryption operates on a page basis. Make this section PMD-aligned
71 * to avoid splitting the pages while mapping the section early.
72 *
73 * Note: We use a separate section so that only this section gets
74 * decrypted to avoid exposing more than we wish.
75 */
76#define BSS_DECRYPTED						\
77	. = ALIGN(PMD_SIZE);					\
78	__start_bss_decrypted = .;				\
79	*(.bss..decrypted);					\
80	. = ALIGN(PAGE_SIZE);					\
81	__start_bss_decrypted_unused = .;			\
82	. = ALIGN(PMD_SIZE);					\
83	__end_bss_decrypted = .;				\
84
85#else
86
87#define X86_ALIGN_RODATA_BEGIN
88#define X86_ALIGN_RODATA_END					\
89		. = ALIGN(PAGE_SIZE);				\
90		__end_rodata_aligned = .;
91
92#define ALIGN_ENTRY_TEXT_BEGIN
93#define ALIGN_ENTRY_TEXT_END
94#define BSS_DECRYPTED
95
96#endif
97
98PHDRS {
99	text PT_LOAD FLAGS(5);          /* R_E */
100	data PT_LOAD FLAGS(6);          /* RW_ */
101#ifdef CONFIG_X86_64
102#ifdef CONFIG_SMP
103	percpu PT_LOAD FLAGS(6);        /* RW_ */
104#endif
105	init PT_LOAD FLAGS(7);          /* RWE */
106#endif
107	note PT_NOTE FLAGS(0);          /* ___ */
108}
109
110SECTIONS
111{
112#ifdef CONFIG_X86_32
113	. = LOAD_OFFSET + LOAD_PHYSICAL_ADDR;
114	phys_startup_32 = ABSOLUTE(startup_32 - LOAD_OFFSET);
115#else
116	. = __START_KERNEL;
117	phys_startup_64 = ABSOLUTE(startup_64 - LOAD_OFFSET);
118#endif
119
120	/* Text and read-only data */
121	.text :  AT(ADDR(.text) - LOAD_OFFSET) {
122		_text = .;
123		_stext = .;
124		/* bootstrapping code */
125		HEAD_TEXT
126		TEXT_TEXT
127		SCHED_TEXT
128		CPUIDLE_TEXT
129		LOCK_TEXT
130		KPROBES_TEXT
131		ALIGN_ENTRY_TEXT_BEGIN
132		ENTRY_TEXT
133		IRQENTRY_TEXT
134		ALIGN_ENTRY_TEXT_END
135		SOFTIRQENTRY_TEXT
136		*(.fixup)
137		*(.gnu.warning)
138
139#ifdef CONFIG_RETPOLINE
140		__indirect_thunk_start = .;
141		*(.text.__x86.indirect_thunk)
142		__indirect_thunk_end = .;
143#endif
144
145		/* End of text section */
146		_etext = .;
147	} :text = 0x9090
148
149	NOTES :text :note
150
151	EXCEPTION_TABLE(16) :text = 0x9090
152
153	/* .text should occupy whole number of pages */
154	. = ALIGN(PAGE_SIZE);
155	X86_ALIGN_RODATA_BEGIN
156	RO_DATA(PAGE_SIZE)
157	X86_ALIGN_RODATA_END
158
159	/* Data */
160	.data : AT(ADDR(.data) - LOAD_OFFSET) {
161		/* Start of data section */
162		_sdata = .;
163
164		/* init_task */
165		INIT_TASK_DATA(THREAD_SIZE)
166
167#ifdef CONFIG_X86_32
168		/* 32 bit has nosave before _edata */
169		NOSAVE_DATA
170#endif
171
172		PAGE_ALIGNED_DATA(PAGE_SIZE)
173
174		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
175
176		DATA_DATA
177		CONSTRUCTORS
178
179		/* rarely changed data like cpu maps */
180		READ_MOSTLY_DATA(INTERNODE_CACHE_BYTES)
181
182		/* End of data section */
183		_edata = .;
184	} :data
185
186	BUG_TABLE
187
188	ORC_UNWIND_TABLE
189
190	. = ALIGN(PAGE_SIZE);
191	__vvar_page = .;
192
193	.vvar : AT(ADDR(.vvar) - LOAD_OFFSET) {
194		/* work around gold bug 13023 */
195		__vvar_beginning_hack = .;
196
197		/* Place all vvars at the offsets in asm/vvar.h. */
198#define EMIT_VVAR(name, offset) 			\
199		. = __vvar_beginning_hack + offset;	\
200		*(.vvar_ ## name)
201#define __VVAR_KERNEL_LDS
202#include <asm/vvar.h>
203#undef __VVAR_KERNEL_LDS
204#undef EMIT_VVAR
205
206		/*
207		 * Pad the rest of the page with zeros.  Otherwise the loader
208		 * can leave garbage here.
209		 */
210		. = __vvar_beginning_hack + PAGE_SIZE;
211	} :data
212
213	. = ALIGN(__vvar_page + PAGE_SIZE, PAGE_SIZE);
214
215	/* Init code and data - will be freed after init */
216	. = ALIGN(PAGE_SIZE);
217	.init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) {
218		__init_begin = .; /* paired with __init_end */
219	}
220
221#if defined(CONFIG_X86_64) && defined(CONFIG_SMP)
222	/*
223	 * percpu offsets are zero-based on SMP.  PERCPU_VADDR() changes the
224	 * output PHDR, so the next output section - .init.text - should
225	 * start another segment - init.
226	 */
227	PERCPU_VADDR(INTERNODE_CACHE_BYTES, 0, :percpu)
228	ASSERT(SIZEOF(.data..percpu) < CONFIG_PHYSICAL_START,
229	       "per-CPU data too large - increase CONFIG_PHYSICAL_START")
230#endif
231
232	INIT_TEXT_SECTION(PAGE_SIZE)
233#ifdef CONFIG_X86_64
234	:init
235#endif
236
237	/*
238	 * Section for code used exclusively before alternatives are run. All
239	 * references to such code must be patched out by alternatives, normally
240	 * by using X86_FEATURE_ALWAYS CPU feature bit.
241	 *
242	 * See static_cpu_has() for an example.
243	 */
244	.altinstr_aux : AT(ADDR(.altinstr_aux) - LOAD_OFFSET) {
245		*(.altinstr_aux)
246	}
247
248	INIT_DATA_SECTION(16)
249
250	.x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
251		__x86_cpu_dev_start = .;
252		*(.x86_cpu_dev.init)
253		__x86_cpu_dev_end = .;
254	}
255
256#ifdef CONFIG_X86_INTEL_MID
257	.x86_intel_mid_dev.init : AT(ADDR(.x86_intel_mid_dev.init) - \
258								LOAD_OFFSET) {
259		__x86_intel_mid_dev_start = .;
260		*(.x86_intel_mid_dev.init)
261		__x86_intel_mid_dev_end = .;
262	}
263#endif
264
265	/*
266	 * start address and size of operations which during runtime
267	 * can be patched with virtualization friendly instructions or
268	 * baremetal native ones. Think page table operations.
269	 * Details in paravirt_types.h
270	 */
271	. = ALIGN(8);
272	.parainstructions : AT(ADDR(.parainstructions) - LOAD_OFFSET) {
273		__parainstructions = .;
274		*(.parainstructions)
275		__parainstructions_end = .;
276	}
277
278	/*
279	 * struct alt_inst entries. From the header (alternative.h):
280	 * "Alternative instructions for different CPU types or capabilities"
281	 * Think locking instructions on spinlocks.
282	 */
283	. = ALIGN(8);
284	.altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) {
285		__alt_instructions = .;
286		*(.altinstructions)
287		__alt_instructions_end = .;
288	}
289
290	/*
291	 * And here are the replacement instructions. The linker sticks
292	 * them as binary blobs. The .altinstructions has enough data to
293	 * get the address and the length of them to patch the kernel safely.
294	 */
295	.altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) {
296		*(.altinstr_replacement)
297	}
298
299	/*
300	 * struct iommu_table_entry entries are injected in this section.
301	 * It is an array of IOMMUs which during run time gets sorted depending
302	 * on its dependency order. After rootfs_initcall is complete
303	 * this section can be safely removed.
304	 */
305	.iommu_table : AT(ADDR(.iommu_table) - LOAD_OFFSET) {
306		__iommu_table = .;
307		*(.iommu_table)
308		__iommu_table_end = .;
309	}
310
311	. = ALIGN(8);
312	.apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) {
313		__apicdrivers = .;
314		*(.apicdrivers);
315		__apicdrivers_end = .;
316	}
317
318	. = ALIGN(8);
319	/*
320	 * .exit.text is discard at runtime, not link time, to deal with
321	 *  references from .altinstructions and .eh_frame
322	 */
323	.exit.text : AT(ADDR(.exit.text) - LOAD_OFFSET) {
324		EXIT_TEXT
325	}
326
327	.exit.data : AT(ADDR(.exit.data) - LOAD_OFFSET) {
328		EXIT_DATA
329	}
330
331#if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP)
332	PERCPU_SECTION(INTERNODE_CACHE_BYTES)
333#endif
334
335	. = ALIGN(PAGE_SIZE);
336
337	/* freed after init ends here */
338	.init.end : AT(ADDR(.init.end) - LOAD_OFFSET) {
339		__init_end = .;
340	}
341
342	/*
343	 * smp_locks might be freed after init
344	 * start/end must be page aligned
345	 */
346	. = ALIGN(PAGE_SIZE);
347	.smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) {
348		__smp_locks = .;
349		*(.smp_locks)
350		. = ALIGN(PAGE_SIZE);
351		__smp_locks_end = .;
352	}
353
354#ifdef CONFIG_X86_64
355	.data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) {
356		NOSAVE_DATA
357	}
358#endif
359
360	/* BSS */
361	. = ALIGN(PAGE_SIZE);
362	.bss : AT(ADDR(.bss) - LOAD_OFFSET) {
363		__bss_start = .;
364		*(.bss..page_aligned)
365		*(BSS_MAIN)
366		BSS_DECRYPTED
367		. = ALIGN(PAGE_SIZE);
368		__bss_stop = .;
369	}
370
371	/*
372	 * The memory occupied from _text to here, __end_of_kernel_reserve, is
373	 * automatically reserved in setup_arch(). Anything after here must be
374	 * explicitly reserved using memblock_reserve() or it will be discarded
375	 * and treated as available memory.
376	 */
377	__end_of_kernel_reserve = .;
378
379	. = ALIGN(PAGE_SIZE);
380	.brk : AT(ADDR(.brk) - LOAD_OFFSET) {
381		__brk_base = .;
382		. += 64 * 1024;		/* 64k alignment slop space */
383		*(.brk_reservation)	/* areas brk users have reserved */
384		__brk_limit = .;
385	}
386
387	. = ALIGN(PAGE_SIZE);		/* keep VO_INIT_SIZE page aligned */
388	_end = .;
389
390#ifdef CONFIG_AMD_MEM_ENCRYPT
391	/*
392	 * Early scratch/workarea section: Lives outside of the kernel proper
393	 * (_text - _end).
394	 *
395	 * Resides after _end because even though the .brk section is after
396	 * __end_of_kernel_reserve, the .brk section is later reserved as a
397	 * part of the kernel. Since it is located after __end_of_kernel_reserve
398	 * it will be discarded and become part of the available memory. As
399	 * such, it can only be used by very early boot code and must not be
400	 * needed afterwards.
401	 *
402	 * Currently used by SME for performing in-place encryption of the
403	 * kernel during boot. Resides on a 2MB boundary to simplify the
404	 * pagetable setup used for SME in-place encryption.
405	 */
406	. = ALIGN(HPAGE_SIZE);
407	.init.scratch : AT(ADDR(.init.scratch) - LOAD_OFFSET) {
408		__init_scratch_begin = .;
409		*(.init.scratch)
410		. = ALIGN(HPAGE_SIZE);
411		__init_scratch_end = .;
412	}
413#endif
414
415	STABS_DEBUG
416	DWARF_DEBUG
417
418	DISCARDS
419	/DISCARD/ : {
420		*(.eh_frame)
421	}
422}
423
424
425#ifdef CONFIG_X86_32
426/*
427 * The ASSERT() sink to . is intentional, for binutils 2.14 compatibility:
428 */
429. = ASSERT((_end - LOAD_OFFSET <= KERNEL_IMAGE_SIZE),
430	   "kernel image bigger than KERNEL_IMAGE_SIZE");
431#else
432/*
433 * Per-cpu symbols which need to be offset from __per_cpu_load
434 * for the boot processor.
435 */
436#define INIT_PER_CPU(x) init_per_cpu__##x = ABSOLUTE(x) + __per_cpu_load
437INIT_PER_CPU(gdt_page);
438INIT_PER_CPU(fixed_percpu_data);
439INIT_PER_CPU(irq_stack_backing_store);
440
441/*
442 * Build-time check on the image size:
443 */
444. = ASSERT((_end - _text <= KERNEL_IMAGE_SIZE),
445	   "kernel image bigger than KERNEL_IMAGE_SIZE");
446
447#ifdef CONFIG_SMP
448. = ASSERT((fixed_percpu_data == 0),
449           "fixed_percpu_data is not at start of per-cpu area");
450#endif
451
452#endif /* CONFIG_X86_32 */
453
454#ifdef CONFIG_KEXEC_CORE
455#include <asm/kexec.h>
456
457. = ASSERT(kexec_control_code_size <= KEXEC_CONTROL_CODE_MAX_SIZE,
458           "kexec control code size is too big");
459#endif
460
461