aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/power
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/power')
-rw-r--r--arch/x86/power/Makefile9
-rw-r--r--arch/x86/power/hibernate_asm_64.S146
-rw-r--r--arch/x86/power/suspend_64.c321
3 files changed, 474 insertions, 2 deletions
diff --git a/arch/x86/power/Makefile b/arch/x86/power/Makefile
index d764ec95006..8ce87fb4abb 100644
--- a/arch/x86/power/Makefile
+++ b/arch/x86/power/Makefile
@@ -1,2 +1,7 @@
1obj-$(CONFIG_PM) += cpu.o 1ifeq ($(CONFIG_X86_64),y)
2obj-$(CONFIG_HIBERNATION) += swsusp.o suspend.o 2 obj-$(CONFIG_PM) += suspend_64.o
3 obj-$(CONFIG_HIBERNATION) += hibernate_asm_64.o
4else
5 obj-$(CONFIG_PM) += cpu.o
6 obj-$(CONFIG_HIBERNATION) += swsusp.o suspend.o
7endif
diff --git a/arch/x86/power/hibernate_asm_64.S b/arch/x86/power/hibernate_asm_64.S
new file mode 100644
index 00000000000..1deb3244b99
--- /dev/null
+++ b/arch/x86/power/hibernate_asm_64.S
@@ -0,0 +1,146 @@
1/*
2 * Hibernation support for x86-64
3 *
4 * Distribute under GPLv2.
5 *
6 * Copyright 2007 Rafael J. Wysocki <rjw@sisk.pl>
7 * Copyright 2005 Andi Kleen <ak@suse.de>
8 * Copyright 2004 Pavel Machek <pavel@suse.cz>
9 *
10 * swsusp_arch_resume must not use any stack or any nonlocal variables while
11 * copying pages:
12 *
13 * Its rewriting one kernel image with another. What is stack in "old"
14 * image could very well be data page in "new" image, and overwriting
15 * your own stack under you is bad idea.
16 */
17
18 .text
19#include <linux/linkage.h>
20#include <asm/segment.h>
21#include <asm/page.h>
22#include <asm/asm-offsets.h>
23
24ENTRY(swsusp_arch_suspend)
25 movq $saved_context, %rax
26 movq %rsp, pt_regs_sp(%rax)
27 movq %rbp, pt_regs_bp(%rax)
28 movq %rsi, pt_regs_si(%rax)
29 movq %rdi, pt_regs_di(%rax)
30 movq %rbx, pt_regs_bx(%rax)
31 movq %rcx, pt_regs_cx(%rax)
32 movq %rdx, pt_regs_dx(%rax)
33 movq %r8, pt_regs_r8(%rax)
34 movq %r9, pt_regs_r9(%rax)
35 movq %r10, pt_regs_r10(%rax)
36 movq %r11, pt_regs_r11(%rax)
37 movq %r12, pt_regs_r12(%rax)
38 movq %r13, pt_regs_r13(%rax)
39 movq %r14, pt_regs_r14(%rax)
40 movq %r15, pt_regs_r15(%rax)
41 pushfq
42 popq pt_regs_flags(%rax)
43
44 /* save the address of restore_registers */
45 movq $restore_registers, %rax
46 movq %rax, restore_jump_address(%rip)
47 /* save cr3 */
48 movq %cr3, %rax
49 movq %rax, restore_cr3(%rip)
50
51 call swsusp_save
52 ret
53
54ENTRY(restore_image)
55 /* switch to temporary page tables */
56 movq $__PAGE_OFFSET, %rdx
57 movq temp_level4_pgt(%rip), %rax
58 subq %rdx, %rax
59 movq %rax, %cr3
60 /* Flush TLB */
61 movq mmu_cr4_features(%rip), %rax
62 movq %rax, %rdx
63 andq $~(1<<7), %rdx # PGE
64 movq %rdx, %cr4; # turn off PGE
65 movq %cr3, %rcx; # flush TLB
66 movq %rcx, %cr3;
67 movq %rax, %cr4; # turn PGE back on
68
69 /* prepare to jump to the image kernel */
70 movq restore_jump_address(%rip), %rax
71 movq restore_cr3(%rip), %rbx
72
73 /* prepare to copy image data to their original locations */
74 movq restore_pblist(%rip), %rdx
75 movq relocated_restore_code(%rip), %rcx
76 jmpq *%rcx
77
78 /* code below has been relocated to a safe page */
79ENTRY(core_restore_code)
80loop:
81 testq %rdx, %rdx
82 jz done
83
84 /* get addresses from the pbe and copy the page */
85 movq pbe_address(%rdx), %rsi
86 movq pbe_orig_address(%rdx), %rdi
87 movq $(PAGE_SIZE >> 3), %rcx
88 rep
89 movsq
90
91 /* progress to the next pbe */
92 movq pbe_next(%rdx), %rdx
93 jmp loop
94done:
95 /* jump to the restore_registers address from the image header */
96 jmpq *%rax
97 /*
98 * NOTE: This assumes that the boot kernel's text mapping covers the
99 * image kernel's page containing restore_registers and the address of
100 * this page is the same as in the image kernel's text mapping (it
101 * should always be true, because the text mapping is linear, starting
102 * from 0, and is supposed to cover the entire kernel text for every
103 * kernel).
104 *
105 * code below belongs to the image kernel
106 */
107
108ENTRY(restore_registers)
109 /* go back to the original page tables */
110 movq %rbx, %cr3
111
112 /* Flush TLB, including "global" things (vmalloc) */
113 movq mmu_cr4_features(%rip), %rax
114 movq %rax, %rdx
115 andq $~(1<<7), %rdx; # PGE
116 movq %rdx, %cr4; # turn off PGE
117 movq %cr3, %rcx; # flush TLB
118 movq %rcx, %cr3
119 movq %rax, %cr4; # turn PGE back on
120
121 /* We don't restore %rax, it must be 0 anyway */
122 movq $saved_context, %rax
123 movq pt_regs_sp(%rax), %rsp
124 movq pt_regs_bp(%rax), %rbp
125 movq pt_regs_si(%rax), %rsi
126 movq pt_regs_di(%rax), %rdi
127 movq pt_regs_bx(%rax), %rbx
128 movq pt_regs_cx(%rax), %rcx
129 movq pt_regs_dx(%rax), %rdx
130 movq pt_regs_r8(%rax), %r8
131 movq pt_regs_r9(%rax), %r9
132 movq pt_regs_r10(%rax), %r10
133 movq pt_regs_r11(%rax), %r11
134 movq pt_regs_r12(%rax), %r12
135 movq pt_regs_r13(%rax), %r13
136 movq pt_regs_r14(%rax), %r14
137 movq pt_regs_r15(%rax), %r15
138 pushq pt_regs_flags(%rax)
139 popfq
140
141 xorq %rax, %rax
142
143 /* tell the hibernation core that we've just restored the memory */
144 movq %rax, in_suspend(%rip)
145
146 ret
diff --git a/arch/x86/power/suspend_64.c b/arch/x86/power/suspend_64.c
new file mode 100644
index 00000000000..d51dbf21d02
--- /dev/null
+++ b/arch/x86/power/suspend_64.c
@@ -0,0 +1,321 @@
1/*
2 * Suspend and hibernation support for x86-64
3 *
4 * Distribute under GPLv2
5 *
6 * Copyright (c) 2007 Rafael J. Wysocki <rjw@sisk.pl>
7 * Copyright (c) 2002 Pavel Machek <pavel@suse.cz>
8 * Copyright (c) 2001 Patrick Mochel <mochel@osdl.org>
9 */
10
11#include <linux/smp.h>
12#include <linux/suspend.h>
13#include <asm/proto.h>
14#include <asm/page.h>
15#include <asm/pgtable.h>
16#include <asm/mtrr.h>
17
18/* References to section boundaries */
19extern const void __nosave_begin, __nosave_end;
20
21static void fix_processor_context(void);
22
23struct saved_context saved_context;
24
25/**
26 * __save_processor_state - save CPU registers before creating a
27 * hibernation image and before restoring the memory state from it
28 * @ctxt - structure to store the registers contents in
29 *
30 * NOTE: If there is a CPU register the modification of which by the
31 * boot kernel (ie. the kernel used for loading the hibernation image)
32 * might affect the operations of the restored target kernel (ie. the one
33 * saved in the hibernation image), then its contents must be saved by this
34 * function. In other words, if kernel A is hibernated and different
35 * kernel B is used for loading the hibernation image into memory, the
36 * kernel A's __save_processor_state() function must save all registers
37 * needed by kernel A, so that it can operate correctly after the resume
38 * regardless of what kernel B does in the meantime.
39 */
40static void __save_processor_state(struct saved_context *ctxt)
41{
42 kernel_fpu_begin();
43
44 /*
45 * descriptor tables
46 */
47 store_gdt((struct desc_ptr *)&ctxt->gdt_limit);
48 store_idt((struct desc_ptr *)&ctxt->idt_limit);
49 store_tr(ctxt->tr);
50
51 /* XMM0..XMM15 should be handled by kernel_fpu_begin(). */
52 /*
53 * segment registers
54 */
55 asm volatile ("movw %%ds, %0" : "=m" (ctxt->ds));
56 asm volatile ("movw %%es, %0" : "=m" (ctxt->es));
57 asm volatile ("movw %%fs, %0" : "=m" (ctxt->fs));
58 asm volatile ("movw %%gs, %0" : "=m" (ctxt->gs));
59 asm volatile ("movw %%ss, %0" : "=m" (ctxt->ss));
60
61 rdmsrl(MSR_FS_BASE, ctxt->fs_base);
62 rdmsrl(MSR_GS_BASE, ctxt->gs_base);
63 rdmsrl(MSR_KERNEL_GS_BASE, ctxt->gs_kernel_base);
64 mtrr_save_fixed_ranges(NULL);
65
66 /*
67 * control registers
68 */
69 rdmsrl(MSR_EFER, ctxt->efer);
70 ctxt->cr0 = read_cr0();
71 ctxt->cr2 = read_cr2();
72 ctxt->cr3 = read_cr3();
73 ctxt->cr4 = read_cr4();
74 ctxt->cr8 = read_cr8();
75}
76
77void save_processor_state(void)
78{
79 __save_processor_state(&saved_context);
80}
81
82static void do_fpu_end(void)
83{
84 /*
85 * Restore FPU regs if necessary
86 */
87 kernel_fpu_end();
88}
89
90/**
91 * __restore_processor_state - restore the contents of CPU registers saved
92 * by __save_processor_state()
93 * @ctxt - structure to load the registers contents from
94 */
95static void __restore_processor_state(struct saved_context *ctxt)
96{
97 /*
98 * control registers
99 */
100 wrmsrl(MSR_EFER, ctxt->efer);
101 write_cr8(ctxt->cr8);
102 write_cr4(ctxt->cr4);
103 write_cr3(ctxt->cr3);
104 write_cr2(ctxt->cr2);
105 write_cr0(ctxt->cr0);
106
107 /*
108 * now restore the descriptor tables to their proper values
109 * ltr is done i fix_processor_context().
110 */
111 load_gdt((const struct desc_ptr *)&ctxt->gdt_limit);
112 load_idt((const struct desc_ptr *)&ctxt->idt_limit);
113
114
115 /*
116 * segment registers
117 */
118 asm volatile ("movw %0, %%ds" :: "r" (ctxt->ds));
119 asm volatile ("movw %0, %%es" :: "r" (ctxt->es));
120 asm volatile ("movw %0, %%fs" :: "r" (ctxt->fs));
121 load_gs_index(ctxt->gs);
122 asm volatile ("movw %0, %%ss" :: "r" (ctxt->ss));
123
124 wrmsrl(MSR_FS_BASE, ctxt->fs_base);
125 wrmsrl(MSR_GS_BASE, ctxt->gs_base);
126 wrmsrl(MSR_KERNEL_GS_BASE, ctxt->gs_kernel_base);
127
128 fix_processor_context();
129
130 do_fpu_end();
131 mtrr_ap_init();
132}
133
134void restore_processor_state(void)
135{
136 __restore_processor_state(&saved_context);
137}
138
139static void fix_processor_context(void)
140{
141 int cpu = smp_processor_id();
142 struct tss_struct *t = &per_cpu(init_tss, cpu);
143
144 /*
145 * This just modifies memory; should not be necessary. But... This
146 * is necessary, because 386 hardware has concept of busy TSS or some
147 * similar stupidity.
148 */
149 set_tss_desc(cpu, t);
150
151 get_cpu_gdt_table(cpu)[GDT_ENTRY_TSS].type = 9;
152
153 syscall_init(); /* This sets MSR_*STAR and related */
154 load_TR_desc(); /* This does ltr */
155 load_LDT(&current->active_mm->context); /* This does lldt */
156
157 /*
158 * Now maybe reload the debug registers
159 */
160 if (current->thread.debugreg7){
161 loaddebug(&current->thread, 0);
162 loaddebug(&current->thread, 1);
163 loaddebug(&current->thread, 2);
164 loaddebug(&current->thread, 3);
165 /* no 4 and 5 */
166 loaddebug(&current->thread, 6);
167 loaddebug(&current->thread, 7);
168 }
169}
170
171#ifdef CONFIG_HIBERNATION
172/* Defined in arch/x86_64/kernel/suspend_asm.S */
173extern int restore_image(void);
174
175/*
176 * Address to jump to in the last phase of restore in order to get to the image
177 * kernel's text (this value is passed in the image header).
178 */
179unsigned long restore_jump_address;
180
181/*
182 * Value of the cr3 register from before the hibernation (this value is passed
183 * in the image header).
184 */
185unsigned long restore_cr3;
186
187pgd_t *temp_level4_pgt;
188
189void *relocated_restore_code;
190
191static int res_phys_pud_init(pud_t *pud, unsigned long address, unsigned long end)
192{
193 long i, j;
194
195 i = pud_index(address);
196 pud = pud + i;
197 for (; i < PTRS_PER_PUD; pud++, i++) {
198 unsigned long paddr;
199 pmd_t *pmd;
200
201 paddr = address + i*PUD_SIZE;
202 if (paddr >= end)
203 break;
204
205 pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
206 if (!pmd)
207 return -ENOMEM;
208 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
209 for (j = 0; j < PTRS_PER_PMD; pmd++, j++, paddr += PMD_SIZE) {
210 unsigned long pe;
211
212 if (paddr >= end)
213 break;
214 pe = __PAGE_KERNEL_LARGE_EXEC | paddr;
215 pe &= __supported_pte_mask;
216 set_pmd(pmd, __pmd(pe));
217 }
218 }
219 return 0;
220}
221
222static int set_up_temporary_mappings(void)
223{
224 unsigned long start, end, next;
225 int error;
226
227 temp_level4_pgt = (pgd_t *)get_safe_page(GFP_ATOMIC);
228 if (!temp_level4_pgt)
229 return -ENOMEM;
230
231 /* It is safe to reuse the original kernel mapping */
232 set_pgd(temp_level4_pgt + pgd_index(__START_KERNEL_map),
233 init_level4_pgt[pgd_index(__START_KERNEL_map)]);
234
235 /* Set up the direct mapping from scratch */
236 start = (unsigned long)pfn_to_kaddr(0);
237 end = (unsigned long)pfn_to_kaddr(end_pfn);
238
239 for (; start < end; start = next) {
240 pud_t *pud = (pud_t *)get_safe_page(GFP_ATOMIC);
241 if (!pud)
242 return -ENOMEM;
243 next = start + PGDIR_SIZE;
244 if (next > end)
245 next = end;
246 if ((error = res_phys_pud_init(pud, __pa(start), __pa(next))))
247 return error;
248 set_pgd(temp_level4_pgt + pgd_index(start),
249 mk_kernel_pgd(__pa(pud)));
250 }
251 return 0;
252}
253
254int swsusp_arch_resume(void)
255{
256 int error;
257
258 /* We have got enough memory and from now on we cannot recover */
259 if ((error = set_up_temporary_mappings()))
260 return error;
261
262 relocated_restore_code = (void *)get_safe_page(GFP_ATOMIC);
263 if (!relocated_restore_code)
264 return -ENOMEM;
265 memcpy(relocated_restore_code, &core_restore_code,
266 &restore_registers - &core_restore_code);
267
268 restore_image();
269 return 0;
270}
271
272/*
273 * pfn_is_nosave - check if given pfn is in the 'nosave' section
274 */
275
276int pfn_is_nosave(unsigned long pfn)
277{
278 unsigned long nosave_begin_pfn = __pa_symbol(&__nosave_begin) >> PAGE_SHIFT;
279 unsigned long nosave_end_pfn = PAGE_ALIGN(__pa_symbol(&__nosave_end)) >> PAGE_SHIFT;
280 return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
281}
282
283struct restore_data_record {
284 unsigned long jump_address;
285 unsigned long cr3;
286 unsigned long magic;
287};
288
289#define RESTORE_MAGIC 0x0123456789ABCDEFUL
290
291/**
292 * arch_hibernation_header_save - populate the architecture specific part
293 * of a hibernation image header
294 * @addr: address to save the data at
295 */
296int arch_hibernation_header_save(void *addr, unsigned int max_size)
297{
298 struct restore_data_record *rdr = addr;
299
300 if (max_size < sizeof(struct restore_data_record))
301 return -EOVERFLOW;
302 rdr->jump_address = restore_jump_address;
303 rdr->cr3 = restore_cr3;
304 rdr->magic = RESTORE_MAGIC;
305 return 0;
306}
307
308/**
309 * arch_hibernation_header_restore - read the architecture specific data
310 * from the hibernation image header
311 * @addr: address to read the data from
312 */
313int arch_hibernation_header_restore(void *addr)
314{
315 struct restore_data_record *rdr = addr;
316
317 restore_jump_address = rdr->jump_address;
318 restore_cr3 = rdr->cr3;
319 return (rdr->magic == RESTORE_MAGIC) ? 0 : -EINVAL;
320}
321#endif /* CONFIG_HIBERNATION */