]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - arch/x86/power/cpu.c
x86/fpu: Uninline kernel_fpu_begin()/end()
[mirror_ubuntu-jammy-kernel.git] / arch / x86 / power / cpu.c
CommitLineData
1da177e4 1/*
6d48becd 2 * Suspend support specific for i386/x86-64.
1da177e4
LT
3 *
4 * Distribute under GPLv2
5 *
cf7700fe 6 * Copyright (c) 2007 Rafael J. Wysocki <rjw@sisk.pl>
a2531293 7 * Copyright (c) 2002 Pavel Machek <pavel@ucw.cz>
1da177e4
LT
8 * Copyright (c) 2001 Patrick Mochel <mochel@osdl.org>
9 */
10
1da177e4 11#include <linux/suspend.h>
69c60c88 12#include <linux/export.h>
f6783d20 13#include <linux/smp.h>
1d9d8639 14#include <linux/perf_event.h>
f6783d20 15
3dd08325 16#include <asm/pgtable.h>
f6783d20 17#include <asm/proto.h>
3ebad590 18#include <asm/mtrr.h>
f6783d20
SL
19#include <asm/page.h>
20#include <asm/mce.h>
a8af7898 21#include <asm/suspend.h>
1e350066 22#include <asm/debugreg.h>
a71c8bc5 23#include <asm/cpu.h>
1da177e4 24
833b2ca0 25#ifdef CONFIG_X86_32
d6efc2f7
AK
26__visible unsigned long saved_context_ebx;
27__visible unsigned long saved_context_esp, saved_context_ebp;
28__visible unsigned long saved_context_esi, saved_context_edi;
29__visible unsigned long saved_context_eflags;
833b2ca0 30#endif
cc456c4e 31struct saved_context saved_context;
1da177e4 32
5c9c9bec
RW
33/**
34 * __save_processor_state - save CPU registers before creating a
35 * hibernation image and before restoring the memory state from it
36 * @ctxt - structure to store the registers contents in
37 *
38 * NOTE: If there is a CPU register the modification of which by the
39 * boot kernel (ie. the kernel used for loading the hibernation image)
40 * might affect the operations of the restored target kernel (ie. the one
41 * saved in the hibernation image), then its contents must be saved by this
42 * function. In other words, if kernel A is hibernated and different
43 * kernel B is used for loading the hibernation image into memory, the
44 * kernel A's __save_processor_state() function must save all registers
45 * needed by kernel A, so that it can operate correctly after the resume
46 * regardless of what kernel B does in the meantime.
47 */
cae45957 48static void __save_processor_state(struct saved_context *ctxt)
1da177e4 49{
f9ebbe53
SL
50#ifdef CONFIG_X86_32
51 mtrr_save_fixed_ranges(NULL);
52#endif
1da177e4
LT
53 kernel_fpu_begin();
54
55 /*
56 * descriptor tables
57 */
f9ebbe53 58#ifdef CONFIG_X86_32
f9ebbe53
SL
59 store_idt(&ctxt->idt);
60#else
61/* CONFIG_X86_64 */
9d1c6e7c 62 store_idt((struct desc_ptr *)&ctxt->idt_limit);
f9ebbe53 63#endif
cc456c4e
KRW
64 /*
65 * We save it here, but restore it only in the hibernate case.
66 * For ACPI S3 resume, this is loaded via 'early_gdt_desc' in 64-bit
67 * mode in "secondary_startup_64". In 32-bit mode it is done via
68 * 'pmode_gdt' in wakeup_start.
69 */
70 ctxt->gdt_desc.size = GDT_SIZE - 1;
71 ctxt->gdt_desc.address = (unsigned long)get_cpu_gdt_table(smp_processor_id());
72
9d1c6e7c 73 store_tr(ctxt->tr);
1da177e4
LT
74
75 /* XMM0..XMM15 should be handled by kernel_fpu_begin(). */
1da177e4
LT
76 /*
77 * segment registers
78 */
f9ebbe53
SL
79#ifdef CONFIG_X86_32
80 savesegment(es, ctxt->es);
81 savesegment(fs, ctxt->fs);
82 savesegment(gs, ctxt->gs);
83 savesegment(ss, ctxt->ss);
84#else
85/* CONFIG_X86_64 */
1da177e4
LT
86 asm volatile ("movw %%ds, %0" : "=m" (ctxt->ds));
87 asm volatile ("movw %%es, %0" : "=m" (ctxt->es));
88 asm volatile ("movw %%fs, %0" : "=m" (ctxt->fs));
89 asm volatile ("movw %%gs, %0" : "=m" (ctxt->gs));
90 asm volatile ("movw %%ss, %0" : "=m" (ctxt->ss));
91
92 rdmsrl(MSR_FS_BASE, ctxt->fs_base);
93 rdmsrl(MSR_GS_BASE, ctxt->gs_base);
94 rdmsrl(MSR_KERNEL_GS_BASE, ctxt->gs_kernel_base);
3ebad590 95 mtrr_save_fixed_ranges(NULL);
1da177e4 96
f9ebbe53
SL
97 rdmsrl(MSR_EFER, ctxt->efer);
98#endif
99
1da177e4 100 /*
cf7700fe 101 * control registers
1da177e4 102 */
f51c9452
GOC
103 ctxt->cr0 = read_cr0();
104 ctxt->cr2 = read_cr2();
105 ctxt->cr3 = read_cr3();
1e02ce4c
AL
106 ctxt->cr4 = __read_cr4_safe();
107#ifdef CONFIG_X86_64
f51c9452 108 ctxt->cr8 = read_cr8();
f9ebbe53 109#endif
85a0e753
OZ
110 ctxt->misc_enable_saved = !rdmsrl_safe(MSR_IA32_MISC_ENABLE,
111 &ctxt->misc_enable);
1da177e4
LT
112}
113
f9ebbe53 114/* Needed by apm.c */
1da177e4
LT
115void save_processor_state(void)
116{
117 __save_processor_state(&saved_context);
b74f05d6 118 x86_platform.save_sched_clock_state();
1da177e4 119}
f9ebbe53
SL
120#ifdef CONFIG_X86_32
121EXPORT_SYMBOL(save_processor_state);
122#endif
1da177e4 123
08967f94 124static void do_fpu_end(void)
1da177e4 125{
08967f94 126 /*
3134d04b 127 * Restore FPU regs if necessary.
08967f94
SL
128 */
129 kernel_fpu_end();
1da177e4
LT
130}
131
3134d04b
SL
132static void fix_processor_context(void)
133{
134 int cpu = smp_processor_id();
24933b82 135 struct tss_struct *t = &per_cpu(cpu_tss, cpu);
4d681be3 136#ifdef CONFIG_X86_64
137 struct desc_struct *desc = get_cpu_gdt_table(cpu);
138 tss_desc tss;
139#endif
3134d04b
SL
140 set_tss_desc(cpu, t); /*
141 * This just modifies memory; should not be
142 * necessary. But... This is necessary, because
143 * 386 hardware has concept of busy TSS or some
144 * similar stupidity.
145 */
146
147#ifdef CONFIG_X86_64
4d681be3 148 memcpy(&tss, &desc[GDT_ENTRY_TSS], sizeof(tss_desc));
149 tss.type = 0x9; /* The available 64-bit TSS (see AMD vol 2, pg 91 */
150 write_gdt_entry(desc, GDT_ENTRY_TSS, &tss, DESC_TSS);
3134d04b
SL
151
152 syscall_init(); /* This sets MSR_*STAR and related */
153#endif
154 load_TR_desc(); /* This does ltr */
155 load_LDT(&current->active_mm->context); /* This does lldt */
9254aaa0
IM
156
157 fpu__resume_cpu();
3134d04b
SL
158}
159
5c9c9bec
RW
160/**
161 * __restore_processor_state - restore the contents of CPU registers saved
162 * by __save_processor_state()
163 * @ctxt - structure to load the registers contents from
164 */
b8f99b3e 165static void notrace __restore_processor_state(struct saved_context *ctxt)
1da177e4 166{
85a0e753
OZ
167 if (ctxt->misc_enable_saved)
168 wrmsrl(MSR_IA32_MISC_ENABLE, ctxt->misc_enable);
1da177e4
LT
169 /*
170 * control registers
171 */
3134d04b
SL
172 /* cr4 was introduced in the Pentium CPU */
173#ifdef CONFIG_X86_32
174 if (ctxt->cr4)
1e02ce4c 175 __write_cr4(ctxt->cr4);
3134d04b
SL
176#else
177/* CONFIG X86_64 */
3c321bce 178 wrmsrl(MSR_EFER, ctxt->efer);
f51c9452 179 write_cr8(ctxt->cr8);
1e02ce4c 180 __write_cr4(ctxt->cr4);
3134d04b 181#endif
f51c9452
GOC
182 write_cr3(ctxt->cr3);
183 write_cr2(ctxt->cr2);
184 write_cr0(ctxt->cr0);
1da177e4 185
8d783b3e
PM
186 /*
187 * now restore the descriptor tables to their proper values
188 * ltr is done i fix_processor_context().
189 */
3134d04b 190#ifdef CONFIG_X86_32
3134d04b
SL
191 load_idt(&ctxt->idt);
192#else
193/* CONFIG_X86_64 */
9d1c6e7c 194 load_idt((const struct desc_ptr *)&ctxt->idt_limit);
3134d04b 195#endif
8d783b3e 196
1da177e4
LT
197 /*
198 * segment registers
199 */
3134d04b
SL
200#ifdef CONFIG_X86_32
201 loadsegment(es, ctxt->es);
202 loadsegment(fs, ctxt->fs);
203 loadsegment(gs, ctxt->gs);
204 loadsegment(ss, ctxt->ss);
205
206 /*
207 * sysenter MSRs
208 */
209 if (boot_cpu_has(X86_FEATURE_SEP))
210 enable_sep_cpu();
211#else
212/* CONFIG_X86_64 */
1da177e4
LT
213 asm volatile ("movw %0, %%ds" :: "r" (ctxt->ds));
214 asm volatile ("movw %0, %%es" :: "r" (ctxt->es));
215 asm volatile ("movw %0, %%fs" :: "r" (ctxt->fs));
216 load_gs_index(ctxt->gs);
217 asm volatile ("movw %0, %%ss" :: "r" (ctxt->ss));
218
219 wrmsrl(MSR_FS_BASE, ctxt->fs_base);
220 wrmsrl(MSR_GS_BASE, ctxt->gs_base);
221 wrmsrl(MSR_KERNEL_GS_BASE, ctxt->gs_kernel_base);
3134d04b 222#endif
1da177e4 223
1da177e4
LT
224 fix_processor_context();
225
226 do_fpu_end();
dba69d10 227 x86_platform.restore_sched_clock_state();
d0af9eed 228 mtrr_bp_restore();
1d9d8639 229 perf_restore_debug_store();
1da177e4
LT
230}
231
3134d04b 232/* Needed by apm.c */
b8f99b3e 233void notrace restore_processor_state(void)
1da177e4
LT
234{
235 __restore_processor_state(&saved_context);
236}
3134d04b
SL
237#ifdef CONFIG_X86_32
238EXPORT_SYMBOL(restore_processor_state);
239#endif
209efae1
FY
240
241/*
242 * When bsp_check() is called in hibernate and suspend, cpu hotplug
243 * is disabled already. So it's unnessary to handle race condition between
244 * cpumask query and cpu hotplug.
245 */
246static int bsp_check(void)
247{
248 if (cpumask_first(cpu_online_mask) != 0) {
249 pr_warn("CPU0 is offline.\n");
250 return -ENODEV;
251 }
252
253 return 0;
254}
255
256static int bsp_pm_callback(struct notifier_block *nb, unsigned long action,
257 void *ptr)
258{
259 int ret = 0;
260
261 switch (action) {
262 case PM_SUSPEND_PREPARE:
263 case PM_HIBERNATION_PREPARE:
264 ret = bsp_check();
265 break;
a71c8bc5
FY
266#ifdef CONFIG_DEBUG_HOTPLUG_CPU0
267 case PM_RESTORE_PREPARE:
268 /*
269 * When system resumes from hibernation, online CPU0 because
270 * 1. it's required for resume and
271 * 2. the CPU was online before hibernation
272 */
273 if (!cpu_online(0))
274 _debug_hotplug_cpu(0, 1);
275 break;
276 case PM_POST_RESTORE:
277 /*
278 * When a resume really happens, this code won't be called.
279 *
280 * This code is called only when user space hibernation software
281 * prepares for snapshot device during boot time. So we just
282 * call _debug_hotplug_cpu() to restore to CPU0's state prior to
283 * preparing the snapshot device.
284 *
285 * This works for normal boot case in our CPU0 hotplug debug
286 * mode, i.e. CPU0 is offline and user mode hibernation
287 * software initializes during boot time.
288 *
289 * If CPU0 is online and user application accesses snapshot
290 * device after boot time, this will offline CPU0 and user may
291 * see different CPU0 state before and after accessing
292 * the snapshot device. But hopefully this is not a case when
293 * user debugging CPU0 hotplug. Even if users hit this case,
294 * they can easily online CPU0 back.
295 *
296 * To simplify this debug code, we only consider normal boot
297 * case. Otherwise we need to remember CPU0's state and restore
298 * to that state and resolve racy conditions etc.
299 */
300 _debug_hotplug_cpu(0, 0);
301 break;
302#endif
209efae1
FY
303 default:
304 break;
305 }
306 return notifier_from_errno(ret);
307}
308
309static int __init bsp_pm_check_init(void)
310{
311 /*
312 * Set this bsp_pm_callback as lower priority than
313 * cpu_hotplug_pm_callback. So cpu_hotplug_pm_callback will be called
314 * earlier to disable cpu hotplug before bsp online check.
315 */
316 pm_notifier(bsp_pm_callback, -INT_MAX);
317 return 0;
318}
319
320core_initcall(bsp_pm_check_init);