1/*
2 * ld script for the x86 kernel
3 *
4 * Historic 32-bit version written by Martin Mares <mj@atrey.karlin.mff.cuni.cz>
5 *
6 * Modernisation, unification and other changes and fixes:
7 *   Copyright (C) 2007-2009  Sam Ravnborg <sam@ravnborg.org>
8 *
9 *
10 * Don't define absolute symbols until and unless you know that symbol
11 * value is should remain constant even if kernel image is relocated
12 * at run time. Absolute symbols are not relocated. If symbol value should
13 * change if kernel is relocated, make the symbol section relative and
14 * put it inside the section definition.
15 */
16
17#ifdef CONFIG_X86_32
18#define LOAD_OFFSET __PAGE_OFFSET
19#else
20#define LOAD_OFFSET __START_KERNEL_map
21#endif
22
23#include <asm-generic/vmlinux.lds.h>
24#include <asm/asm-offsets.h>
25#include <asm/thread_info.h>
26#include <asm/page_types.h>
27#include <asm/cache.h>
28#include <asm/boot.h>
29
30#undef i386     /* in case the preprocessor is a 32bit one */
31
32OUTPUT_FORMAT(CONFIG_OUTPUT_FORMAT, CONFIG_OUTPUT_FORMAT, CONFIG_OUTPUT_FORMAT)
33
34#ifdef CONFIG_X86_32
35OUTPUT_ARCH(i386)
36ENTRY(phys_startup_32)
37jiffies = jiffies_64;
38#else
39OUTPUT_ARCH(i386:x86-64)
40ENTRY(phys_startup_64)
41jiffies_64 = jiffies;
42#endif
43
44#if defined(CONFIG_X86_64) && defined(CONFIG_DEBUG_RODATA)
45/*
46 * On 64-bit, align RODATA to 2MB so that even with CONFIG_DEBUG_RODATA
47 * we retain large page mappings for boundaries spanning kernel text, rodata
48 * and data sections.
49 *
50 * However, kernel identity mappings will have different RWX permissions
51 * to the pages mapping to text and to the pages padding (which are freed) the
52 * text section. Hence kernel identity mappings will be broken to smaller
53 * pages. For 64-bit, kernel text and kernel identity mappings are different,
54 * so we can enable protection checks that come with CONFIG_DEBUG_RODATA,
55 * as well as retain 2MB large page mappings for kernel text.
56 */
57#define X64_ALIGN_DEBUG_RODATA_BEGIN	. = ALIGN(HPAGE_SIZE);
58
59#define X64_ALIGN_DEBUG_RODATA_END				\
60		. = ALIGN(HPAGE_SIZE);				\
61		__end_rodata_hpage_align = .;
62
63#else
64
65#define X64_ALIGN_DEBUG_RODATA_BEGIN
66#define X64_ALIGN_DEBUG_RODATA_END
67
68#endif
69
70PHDRS {
71	text PT_LOAD FLAGS(5);          /* R_E */
72	data PT_LOAD FLAGS(6);          /* RW_ */
73#ifdef CONFIG_X86_64
74	user PT_LOAD FLAGS(5);          /* R_E */
75#ifdef CONFIG_SMP
76	percpu PT_LOAD FLAGS(6);        /* RW_ */
77#endif
78	init PT_LOAD FLAGS(7);          /* RWE */
79#endif
80	note PT_NOTE FLAGS(0);          /* ___ */
81}
82
83SECTIONS
84{
85#ifdef CONFIG_X86_32
86        . = LOAD_OFFSET + LOAD_PHYSICAL_ADDR;
87        phys_startup_32 = startup_32 - LOAD_OFFSET;
88#else
89        . = __START_KERNEL;
90        phys_startup_64 = startup_64 - LOAD_OFFSET;
91#endif
92
93	/* Text and read-only data */
94	.text :  AT(ADDR(.text) - LOAD_OFFSET) {
95		_text = .;
96		/* bootstrapping code */
97		HEAD_TEXT
98#ifdef CONFIG_X86_32
99		. = ALIGN(PAGE_SIZE);
100		*(.text..page_aligned)
101#endif
102		. = ALIGN(8);
103		_stext = .;
104		TEXT_TEXT
105		SCHED_TEXT
106		LOCK_TEXT
107		KPROBES_TEXT
108		ENTRY_TEXT
109		IRQENTRY_TEXT
110		*(.fixup)
111		*(.gnu.warning)
112		/* End of text section */
113		_etext = .;
114	} :text = 0x9090
115
116	NOTES :text :note
117
118	EXCEPTION_TABLE(16) :text = 0x9090
119
120#if defined(CONFIG_DEBUG_RODATA)
121	/* .text should occupy whole number of pages */
122	. = ALIGN(PAGE_SIZE);
123#endif
124	X64_ALIGN_DEBUG_RODATA_BEGIN
125	RO_DATA(PAGE_SIZE)
126	X64_ALIGN_DEBUG_RODATA_END
127
128	/* Data */
129	.data : AT(ADDR(.data) - LOAD_OFFSET) {
130		/* Start of data section */
131		_sdata = .;
132
133		/* init_task */
134		INIT_TASK_DATA(THREAD_SIZE)
135
136#ifdef CONFIG_X86_32
137		/* 32 bit has nosave before _edata */
138		NOSAVE_DATA
139#endif
140
141		PAGE_ALIGNED_DATA(PAGE_SIZE)
142
143		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
144
145		DATA_DATA
146		CONSTRUCTORS
147
148		/* rarely changed data like cpu maps */
149		READ_MOSTLY_DATA(INTERNODE_CACHE_BYTES)
150
151		/* End of data section */
152		_edata = .;
153	} :data
154
155#ifdef CONFIG_X86_64
156
157#define VSYSCALL_ADDR (-10*1024*1024)
158
159#define VLOAD_OFFSET (VSYSCALL_ADDR - __vsyscall_0 + LOAD_OFFSET)
160#define VLOAD(x) (ADDR(x) - VLOAD_OFFSET)
161
162#define VVIRT_OFFSET (VSYSCALL_ADDR - __vsyscall_0)
163#define VVIRT(x) (ADDR(x) - VVIRT_OFFSET)
164
165	. = ALIGN(4096);
166	__vsyscall_0 = .;
167
168	. = VSYSCALL_ADDR;
169	.vsyscall_0 : AT(VLOAD(.vsyscall_0)) {
170		*(.vsyscall_0)
171	} :user
172
173	. = ALIGN(L1_CACHE_BYTES);
174	.vsyscall_fn : AT(VLOAD(.vsyscall_fn)) {
175		*(.vsyscall_fn)
176	}
177
178	. = ALIGN(L1_CACHE_BYTES);
179	.vsyscall_gtod_data : AT(VLOAD(.vsyscall_gtod_data)) {
180		*(.vsyscall_gtod_data)
181	}
182
183	vsyscall_gtod_data = VVIRT(.vsyscall_gtod_data);
184	.vsyscall_clock : AT(VLOAD(.vsyscall_clock)) {
185		*(.vsyscall_clock)
186	}
187	vsyscall_clock = VVIRT(.vsyscall_clock);
188
189
190	.vsyscall_1 ADDR(.vsyscall_0) + 1024: AT(VLOAD(.vsyscall_1)) {
191		*(.vsyscall_1)
192	}
193	.vsyscall_2 ADDR(.vsyscall_0) + 2048: AT(VLOAD(.vsyscall_2)) {
194		*(.vsyscall_2)
195	}
196
197	.vgetcpu_mode : AT(VLOAD(.vgetcpu_mode)) {
198		*(.vgetcpu_mode)
199	}
200	vgetcpu_mode = VVIRT(.vgetcpu_mode);
201
202	. = ALIGN(L1_CACHE_BYTES);
203	.jiffies : AT(VLOAD(.jiffies)) {
204		*(.jiffies)
205	}
206	jiffies = VVIRT(.jiffies);
207
208	.vsyscall_3 ADDR(.vsyscall_0) + 3072: AT(VLOAD(.vsyscall_3)) {
209		*(.vsyscall_3)
210	}
211
212	. = __vsyscall_0 + PAGE_SIZE;
213
214#undef VSYSCALL_ADDR
215#undef VLOAD_OFFSET
216#undef VLOAD
217#undef VVIRT_OFFSET
218#undef VVIRT
219
220#endif /* CONFIG_X86_64 */
221
222	/* Init code and data - will be freed after init */
223	. = ALIGN(PAGE_SIZE);
224	.init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) {
225		__init_begin = .; /* paired with __init_end */
226	}
227
228#if defined(CONFIG_X86_64) && defined(CONFIG_SMP)
229	/*
230	 * percpu offsets are zero-based on SMP.  PERCPU_VADDR() changes the
231	 * output PHDR, so the next output section - .init.text - should
232	 * start another segment - init.
233	 */
234	PERCPU_VADDR(INTERNODE_CACHE_BYTES, 0, :percpu)
235#endif
236
237	INIT_TEXT_SECTION(PAGE_SIZE)
238#ifdef CONFIG_X86_64
239	:init
240#endif
241
242	INIT_DATA_SECTION(16)
243
244	/*
245	 * Code and data for a variety of lowlevel trampolines, to be
246	 * copied into base memory (< 1 MiB) during initialization.
247	 * Since it is copied early, the main copy can be discarded
248	 * afterwards.
249	 */
250	 .x86_trampoline : AT(ADDR(.x86_trampoline) - LOAD_OFFSET) {
251		x86_trampoline_start = .;
252		*(.x86_trampoline)
253		x86_trampoline_end = .;
254	}
255
256	.x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
257		__x86_cpu_dev_start = .;
258		*(.x86_cpu_dev.init)
259		__x86_cpu_dev_end = .;
260	}
261
262	/*
263	 * start address and size of operations which during runtime
264	 * can be patched with virtualization friendly instructions or
265	 * baremetal native ones. Think page table operations.
266	 * Details in paravirt_types.h
267	 */
268	. = ALIGN(8);
269	.parainstructions : AT(ADDR(.parainstructions) - LOAD_OFFSET) {
270		__parainstructions = .;
271		*(.parainstructions)
272		__parainstructions_end = .;
273	}
274
275	/*
276	 * struct alt_inst entries. From the header (alternative.h):
277	 * "Alternative instructions for different CPU types or capabilities"
278	 * Think locking instructions on spinlocks.
279	 */
280	. = ALIGN(8);
281	.altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) {
282		__alt_instructions = .;
283		*(.altinstructions)
284		__alt_instructions_end = .;
285	}
286
287	/*
288	 * And here are the replacement instructions. The linker sticks
289	 * them as binary blobs. The .altinstructions has enough data to
290	 * get the address and the length of them to patch the kernel safely.
291	 */
292	.altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) {
293		*(.altinstr_replacement)
294	}
295
296	/*
297	 * struct iommu_table_entry entries are injected in this section.
298	 * It is an array of IOMMUs which during run time gets sorted depending
299	 * on its dependency order. After rootfs_initcall is complete
300	 * this section can be safely removed.
301	 */
302	.iommu_table : AT(ADDR(.iommu_table) - LOAD_OFFSET) {
303		__iommu_table = .;
304		*(.iommu_table)
305		__iommu_table_end = .;
306	}
307
308	. = ALIGN(8);
309	/*
310	 * .exit.text is discard at runtime, not link time, to deal with
311	 *  references from .altinstructions and .eh_frame
312	 */
313	.exit.text : AT(ADDR(.exit.text) - LOAD_OFFSET) {
314		EXIT_TEXT
315	}
316
317	.exit.data : AT(ADDR(.exit.data) - LOAD_OFFSET) {
318		EXIT_DATA
319	}
320
321#if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP)
322	PERCPU(INTERNODE_CACHE_BYTES, PAGE_SIZE)
323#endif
324
325	. = ALIGN(PAGE_SIZE);
326
327	/* freed after init ends here */
328	.init.end : AT(ADDR(.init.end) - LOAD_OFFSET) {
329		__init_end = .;
330	}
331
332	/*
333	 * smp_locks might be freed after init
334	 * start/end must be page aligned
335	 */
336	. = ALIGN(PAGE_SIZE);
337	.smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) {
338		__smp_locks = .;
339		*(.smp_locks)
340		. = ALIGN(PAGE_SIZE);
341		__smp_locks_end = .;
342	}
343
344#ifdef CONFIG_X86_64
345	.data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) {
346		NOSAVE_DATA
347	}
348#endif
349
350	/* BSS */
351	. = ALIGN(PAGE_SIZE);
352	.bss : AT(ADDR(.bss) - LOAD_OFFSET) {
353		__bss_start = .;
354		*(.bss..page_aligned)
355		*(.bss)
356		. = ALIGN(PAGE_SIZE);
357		__bss_stop = .;
358	}
359
360	. = ALIGN(PAGE_SIZE);
361	.brk : AT(ADDR(.brk) - LOAD_OFFSET) {
362		__brk_base = .;
363		. += 64 * 1024;		/* 64k alignment slop space */
364		*(.brk_reservation)	/* areas brk users have reserved */
365		__brk_limit = .;
366	}
367
368	_end = .;
369
370        STABS_DEBUG
371        DWARF_DEBUG
372
373	/* Sections to be discarded */
374	DISCARDS
375	/DISCARD/ : { *(.eh_frame) }
376}
377
378
379#ifdef CONFIG_X86_32
380/*
381 * The ASSERT() sink to . is intentional, for binutils 2.14 compatibility:
382 */
383. = ASSERT((_end - LOAD_OFFSET <= KERNEL_IMAGE_SIZE),
384	   "kernel image bigger than KERNEL_IMAGE_SIZE");
385#else
386/*
387 * Per-cpu symbols which need to be offset from __per_cpu_load
388 * for the boot processor.
389 */
390#define INIT_PER_CPU(x) init_per_cpu__##x = x + __per_cpu_load
391INIT_PER_CPU(gdt_page);
392INIT_PER_CPU(irq_stack_union);
393
394/*
395 * Build-time check on the image size:
396 */
397. = ASSERT((_end - _text <= KERNEL_IMAGE_SIZE),
398	   "kernel image bigger than KERNEL_IMAGE_SIZE");
399
400#ifdef CONFIG_SMP
401. = ASSERT((irq_stack_union == 0),
402           "irq_stack_union is not at start of per-cpu area");
403#endif
404
405#endif /* CONFIG_X86_32 */
406
407#ifdef CONFIG_KEXEC
408#include <asm/kexec.h>
409
410. = ASSERT(kexec_control_code_size <= KEXEC_CONTROL_CODE_MAX_SIZE,
411           "kexec control code size is too big");
412#endif
413
414