2 * ld script for the x86 kernel
4 * Historic 32-bit version written by Martin Mares <mj@atrey.karlin.mff.cuni.cz>
6 * Modernisation, unification and other changes and fixes:
7 * Copyright (C) 2007-2009 Sam Ravnborg <sam@ravnborg.org>
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.
18 #define LOAD_OFFSET __PAGE_OFFSET
20 #define LOAD_OFFSET __START_KERNEL_map
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>
30 #undef i386 /* in case the preprocessor is a 32bit one */
32 OUTPUT_FORMAT(CONFIG_OUTPUT_FORMAT, CONFIG_OUTPUT_FORMAT, CONFIG_OUTPUT_FORMAT)
36 ENTRY(phys_startup_32)
39 OUTPUT_ARCH(i386:x86-64)
40 ENTRY(phys_startup_64)
44 #if defined(CONFIG_X86_64) && defined(CONFIG_DEBUG_RODATA)
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
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.
57 #define X64_ALIGN_DEBUG_RODATA_BEGIN . = ALIGN(HPAGE_SIZE);
59 #define X64_ALIGN_DEBUG_RODATA_END \
60 . = ALIGN(HPAGE_SIZE); \
61 __end_rodata_hpage_align = .;
65 #define X64_ALIGN_DEBUG_RODATA_BEGIN
66 #define X64_ALIGN_DEBUG_RODATA_END
71 text PT_LOAD FLAGS(5); /* R_E */
72 data PT_LOAD FLAGS(6); /* RW_ */
74 user PT_LOAD FLAGS(5); /* R_E */
76 percpu PT_LOAD FLAGS(6); /* RW_ */
78 init PT_LOAD FLAGS(7); /* RWE */
80 note PT_NOTE FLAGS(0); /* ___ */
86 . = LOAD_OFFSET + LOAD_PHYSICAL_ADDR;
87 phys_startup_32 = startup_32 - LOAD_OFFSET;
90 phys_startup_64 = startup_64 - LOAD_OFFSET;
93 /* Text and read-only data */
94 .text : AT(ADDR(.text) - LOAD_OFFSET) {
96 /* bootstrapping code */
100 *(.text..page_aligned)
112 /* End of text section */
118 EXCEPTION_TABLE(16) :text = 0x9090
120 #if defined(CONFIG_DEBUG_RODATA)
121 /* .text should occupy whole number of pages */
122 . = ALIGN(PAGE_SIZE);
124 X64_ALIGN_DEBUG_RODATA_BEGIN
126 X64_ALIGN_DEBUG_RODATA_END
129 .data : AT(ADDR(.data) - LOAD_OFFSET) {
130 /* Start of data section */
134 INIT_TASK_DATA(THREAD_SIZE)
137 /* 32 bit has nosave before _edata */
141 PAGE_ALIGNED_DATA(PAGE_SIZE)
143 CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
148 /* rarely changed data like cpu maps */
149 READ_MOSTLY_DATA(INTERNODE_CACHE_BYTES)
151 /* End of data section */
157 #define VSYSCALL_ADDR (-10*1024*1024)
159 #define VLOAD_OFFSET (VSYSCALL_ADDR - __vsyscall_0 + LOAD_OFFSET)
160 #define VLOAD(x) (ADDR(x) - VLOAD_OFFSET)
162 #define VVIRT_OFFSET (VSYSCALL_ADDR - __vsyscall_0)
163 #define VVIRT(x) (ADDR(x) - VVIRT_OFFSET)
164 #define EMIT_VVAR(x, offset) .vsyscall_var_ ## x \
165 ADDR(.vsyscall_0) + offset \
166 : AT(VLOAD(.vsyscall_var_ ## x)) { \
167 *(.vsyscall_var_ ## x) \
169 x = VVIRT(.vsyscall_var_ ## x);
175 .vsyscall_0 : AT(VLOAD(.vsyscall_0)) {
179 . = ALIGN(L1_CACHE_BYTES);
180 .vsyscall_fn : AT(VLOAD(.vsyscall_fn)) {
184 .vsyscall_1 ADDR(.vsyscall_0) + 1024: AT(VLOAD(.vsyscall_1)) {
187 .vsyscall_2 ADDR(.vsyscall_0) + 2048: AT(VLOAD(.vsyscall_2)) {
191 .vsyscall_3 ADDR(.vsyscall_0) + 3072: AT(VLOAD(.vsyscall_3)) {
195 #define __VVAR_KERNEL_LDS
196 #include <asm/vvar.h>
197 #undef __VVAR_KERNEL_LDS
199 . = __vsyscall_0 + PAGE_SIZE;
208 #endif /* CONFIG_X86_64 */
210 /* Init code and data - will be freed after init */
211 . = ALIGN(PAGE_SIZE);
212 .init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) {
213 __init_begin = .; /* paired with __init_end */
216 #if defined(CONFIG_X86_64) && defined(CONFIG_SMP)
218 * percpu offsets are zero-based on SMP. PERCPU_VADDR() changes the
219 * output PHDR, so the next output section - .init.text - should
220 * start another segment - init.
222 PERCPU_VADDR(INTERNODE_CACHE_BYTES, 0, :percpu)
225 INIT_TEXT_SECTION(PAGE_SIZE)
230 INIT_DATA_SECTION(16)
233 * Code and data for a variety of lowlevel trampolines, to be
234 * copied into base memory (< 1 MiB) during initialization.
235 * Since it is copied early, the main copy can be discarded
238 .x86_trampoline : AT(ADDR(.x86_trampoline) - LOAD_OFFSET) {
239 x86_trampoline_start = .;
241 x86_trampoline_end = .;
244 .x86_cpu_dev.init : AT(ADDR(.x86_cpu_dev.init) - LOAD_OFFSET) {
245 __x86_cpu_dev_start = .;
247 __x86_cpu_dev_end = .;
251 * start address and size of operations which during runtime
252 * can be patched with virtualization friendly instructions or
253 * baremetal native ones. Think page table operations.
254 * Details in paravirt_types.h
257 .parainstructions : AT(ADDR(.parainstructions) - LOAD_OFFSET) {
258 __parainstructions = .;
260 __parainstructions_end = .;
264 * struct alt_inst entries. From the header (alternative.h):
265 * "Alternative instructions for different CPU types or capabilities"
266 * Think locking instructions on spinlocks.
269 .altinstructions : AT(ADDR(.altinstructions) - LOAD_OFFSET) {
270 __alt_instructions = .;
272 __alt_instructions_end = .;
276 * And here are the replacement instructions. The linker sticks
277 * them as binary blobs. The .altinstructions has enough data to
278 * get the address and the length of them to patch the kernel safely.
280 .altinstr_replacement : AT(ADDR(.altinstr_replacement) - LOAD_OFFSET) {
281 *(.altinstr_replacement)
285 * struct iommu_table_entry entries are injected in this section.
286 * It is an array of IOMMUs which during run time gets sorted depending
287 * on its dependency order. After rootfs_initcall is complete
288 * this section can be safely removed.
290 .iommu_table : AT(ADDR(.iommu_table) - LOAD_OFFSET) {
293 __iommu_table_end = .;
297 .apicdrivers : AT(ADDR(.apicdrivers) - LOAD_OFFSET) {
300 __apicdrivers_end = .;
305 * .exit.text is discard at runtime, not link time, to deal with
306 * references from .altinstructions and .eh_frame
308 .exit.text : AT(ADDR(.exit.text) - LOAD_OFFSET) {
312 .exit.data : AT(ADDR(.exit.data) - LOAD_OFFSET) {
316 #if !defined(CONFIG_X86_64) || !defined(CONFIG_SMP)
317 PERCPU_SECTION(INTERNODE_CACHE_BYTES)
320 . = ALIGN(PAGE_SIZE);
322 /* freed after init ends here */
323 .init.end : AT(ADDR(.init.end) - LOAD_OFFSET) {
328 * smp_locks might be freed after init
329 * start/end must be page aligned
331 . = ALIGN(PAGE_SIZE);
332 .smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) {
335 . = ALIGN(PAGE_SIZE);
340 .data_nosave : AT(ADDR(.data_nosave) - LOAD_OFFSET) {
346 . = ALIGN(PAGE_SIZE);
347 .bss : AT(ADDR(.bss) - LOAD_OFFSET) {
349 *(.bss..page_aligned)
351 . = ALIGN(PAGE_SIZE);
355 . = ALIGN(PAGE_SIZE);
356 .brk : AT(ADDR(.brk) - LOAD_OFFSET) {
358 . += 64 * 1024; /* 64k alignment slop space */
359 *(.brk_reservation) /* areas brk users have reserved */
368 /* Sections to be discarded */
370 /DISCARD/ : { *(.eh_frame) }
376 * The ASSERT() sink to . is intentional, for binutils 2.14 compatibility:
378 . = ASSERT((_end - LOAD_OFFSET <= KERNEL_IMAGE_SIZE),
379 "kernel image bigger than KERNEL_IMAGE_SIZE");
382 * Per-cpu symbols which need to be offset from __per_cpu_load
383 * for the boot processor.
385 #define INIT_PER_CPU(x) init_per_cpu__##x = x + __per_cpu_load
386 INIT_PER_CPU(gdt_page);
387 INIT_PER_CPU(irq_stack_union);
390 * Build-time check on the image size:
392 . = ASSERT((_end - _text <= KERNEL_IMAGE_SIZE),
393 "kernel image bigger than KERNEL_IMAGE_SIZE");
396 . = ASSERT((irq_stack_union == 0),
397 "irq_stack_union is not at start of per-cpu area");
400 #endif /* CONFIG_X86_32 */
403 #include <asm/kexec.h>
405 . = ASSERT(kexec_control_code_size <= KEXEC_CONTROL_CODE_MAX_SIZE,
406 "kexec control code size is too big");