]>
Commit | Line | Data |
---|---|---|
790c73f6 GOC |
1 | /* KVM paravirtual clock driver. A clocksource implementation |
2 | Copyright (C) 2008 Glauber de Oliveira Costa, Red Hat Inc. | |
3 | ||
4 | This program is free software; you can redistribute it and/or modify | |
5 | it under the terms of the GNU General Public License as published by | |
6 | the Free Software Foundation; either version 2 of the License, or | |
7 | (at your option) any later version. | |
8 | ||
9 | This program is distributed in the hope that it will be useful, | |
10 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | GNU General Public License for more details. | |
13 | ||
14 | You should have received a copy of the GNU General Public License | |
15 | along with this program; if not, write to the Free Software | |
16 | Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
17 | */ | |
18 | ||
19 | #include <linux/clocksource.h> | |
20 | #include <linux/kvm_para.h> | |
f6e16d5a | 21 | #include <asm/pvclock.h> |
790c73f6 GOC |
22 | #include <asm/msr.h> |
23 | #include <asm/apic.h> | |
24 | #include <linux/percpu.h> | |
3b5d56b9 | 25 | #include <linux/hardirq.h> |
7069ed67 | 26 | #include <linux/memblock.h> |
0ad83caa | 27 | #include <linux/sched.h> |
736decac TG |
28 | |
29 | #include <asm/x86_init.h> | |
1e977aa1 | 30 | #include <asm/reboot.h> |
790c73f6 | 31 | |
404f6aac | 32 | static int kvmclock __ro_after_init = 1; |
838815a7 GC |
33 | static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME; |
34 | static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK; | |
a5a1d1c2 | 35 | static u64 kvm_sched_clock_offset; |
790c73f6 GOC |
36 | |
37 | static int parse_no_kvmclock(char *arg) | |
38 | { | |
39 | kvmclock = 0; | |
40 | return 0; | |
41 | } | |
42 | early_param("no-kvmclock", parse_no_kvmclock); | |
43 | ||
44 | /* The hypervisor will put information about time periodically here */ | |
3dc4f7cf | 45 | static struct pvclock_vsyscall_time_info *hv_clock; |
f6e16d5a | 46 | static struct pvclock_wall_clock wall_clock; |
790c73f6 | 47 | |
dac16fba AL |
48 | struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void) |
49 | { | |
50 | return hv_clock; | |
51 | } | |
52 | ||
790c73f6 GOC |
53 | /* |
54 | * The wallclock is the time of day when we booted. Since then, some time may | |
55 | * have elapsed since the hypervisor wrote the data. So we try to account for | |
56 | * that with system time | |
57 | */ | |
3565184e | 58 | static void kvm_get_wallclock(struct timespec *now) |
790c73f6 | 59 | { |
f6e16d5a | 60 | struct pvclock_vcpu_time_info *vcpu_time; |
790c73f6 | 61 | int low, high; |
7069ed67 | 62 | int cpu; |
790c73f6 | 63 | |
a20316d2 GC |
64 | low = (int)__pa_symbol(&wall_clock); |
65 | high = ((u64)__pa_symbol(&wall_clock) >> 32); | |
838815a7 GC |
66 | |
67 | native_write_msr(msr_kvm_wall_clock, low, high); | |
790c73f6 | 68 | |
c6338ce4 | 69 | cpu = get_cpu(); |
7069ed67 | 70 | |
3dc4f7cf | 71 | vcpu_time = &hv_clock[cpu].pvti; |
3565184e | 72 | pvclock_read_wallclock(&wall_clock, vcpu_time, now); |
7069ed67 | 73 | |
c6338ce4 | 74 | put_cpu(); |
790c73f6 GOC |
75 | } |
76 | ||
3565184e | 77 | static int kvm_set_wallclock(const struct timespec *now) |
790c73f6 | 78 | { |
f6e16d5a | 79 | return -1; |
790c73f6 GOC |
80 | } |
81 | ||
a5a1d1c2 | 82 | static u64 kvm_clock_read(void) |
790c73f6 | 83 | { |
f6e16d5a | 84 | struct pvclock_vcpu_time_info *src; |
a5a1d1c2 | 85 | u64 ret; |
7069ed67 | 86 | int cpu; |
790c73f6 | 87 | |
95ef1e52 | 88 | preempt_disable_notrace(); |
7069ed67 | 89 | cpu = smp_processor_id(); |
3dc4f7cf | 90 | src = &hv_clock[cpu].pvti; |
f6e16d5a | 91 | ret = pvclock_clocksource_read(src); |
95ef1e52 | 92 | preempt_enable_notrace(); |
f6e16d5a | 93 | return ret; |
790c73f6 | 94 | } |
f6e16d5a | 95 | |
a5a1d1c2 | 96 | static u64 kvm_clock_get_cycles(struct clocksource *cs) |
8e19608e MD |
97 | { |
98 | return kvm_clock_read(); | |
99 | } | |
100 | ||
a5a1d1c2 | 101 | static u64 kvm_sched_clock_read(void) |
72c930dc RK |
102 | { |
103 | return kvm_clock_read() - kvm_sched_clock_offset; | |
104 | } | |
105 | ||
106 | static inline void kvm_sched_clock_init(bool stable) | |
107 | { | |
108 | if (!stable) { | |
109 | pv_time_ops.sched_clock = kvm_clock_read; | |
110 | return; | |
111 | } | |
112 | ||
113 | kvm_sched_clock_offset = kvm_clock_read(); | |
114 | pv_time_ops.sched_clock = kvm_sched_clock_read; | |
115 | set_sched_clock_stable(); | |
116 | ||
117 | printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n", | |
118 | kvm_sched_clock_offset); | |
119 | ||
120 | BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) > | |
121 | sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time)); | |
122 | } | |
123 | ||
0293615f GC |
124 | /* |
125 | * If we don't do that, there is the possibility that the guest | |
126 | * will calibrate under heavy load - thus, getting a lower lpj - | |
127 | * and execute the delays themselves without load. This is wrong, | |
128 | * because no delay loop can finish beforehand. | |
129 | * Any heuristics is subject to fail, because ultimately, a large | |
130 | * poll of guests can be running and trouble each other. So we preset | |
131 | * lpj here | |
132 | */ | |
133 | static unsigned long kvm_get_tsc_khz(void) | |
134 | { | |
e93353c9 | 135 | struct pvclock_vcpu_time_info *src; |
7069ed67 MT |
136 | int cpu; |
137 | unsigned long tsc_khz; | |
138 | ||
c6338ce4 | 139 | cpu = get_cpu(); |
3dc4f7cf | 140 | src = &hv_clock[cpu].pvti; |
7069ed67 | 141 | tsc_khz = pvclock_tsc_khz(src); |
c6338ce4 | 142 | put_cpu(); |
7069ed67 | 143 | return tsc_khz; |
0293615f GC |
144 | } |
145 | ||
146 | static void kvm_get_preset_lpj(void) | |
147 | { | |
0293615f GC |
148 | unsigned long khz; |
149 | u64 lpj; | |
150 | ||
e93353c9 | 151 | khz = kvm_get_tsc_khz(); |
0293615f GC |
152 | |
153 | lpj = ((u64)khz * 1000); | |
154 | do_div(lpj, HZ); | |
155 | preset_lpj = lpj; | |
156 | } | |
157 | ||
3b5d56b9 EM |
158 | bool kvm_check_and_clear_guest_paused(void) |
159 | { | |
160 | bool ret = false; | |
161 | struct pvclock_vcpu_time_info *src; | |
7069ed67 MT |
162 | int cpu = smp_processor_id(); |
163 | ||
164 | if (!hv_clock) | |
165 | return ret; | |
3b5d56b9 | 166 | |
3dc4f7cf | 167 | src = &hv_clock[cpu].pvti; |
3b5d56b9 | 168 | if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) { |
7069ed67 | 169 | src->flags &= ~PVCLOCK_GUEST_STOPPED; |
d63285e9 | 170 | pvclock_touch_watchdogs(); |
3b5d56b9 EM |
171 | ret = true; |
172 | } | |
173 | ||
174 | return ret; | |
175 | } | |
3b5d56b9 | 176 | |
790c73f6 GOC |
177 | static struct clocksource kvm_clock = { |
178 | .name = "kvm-clock", | |
8e19608e | 179 | .read = kvm_clock_get_cycles, |
790c73f6 GOC |
180 | .rating = 400, |
181 | .mask = CLOCKSOURCE_MASK(64), | |
790c73f6 GOC |
182 | .flags = CLOCK_SOURCE_IS_CONTINUOUS, |
183 | }; | |
184 | ||
ca3f1017 | 185 | int kvm_register_clock(char *txt) |
790c73f6 GOC |
186 | { |
187 | int cpu = smp_processor_id(); | |
19b6a85b | 188 | int low, high, ret; |
fe1140cc JK |
189 | struct pvclock_vcpu_time_info *src; |
190 | ||
191 | if (!hv_clock) | |
192 | return 0; | |
19b6a85b | 193 | |
fe1140cc | 194 | src = &hv_clock[cpu].pvti; |
5dfd486c DH |
195 | low = (int)slow_virt_to_phys(src) | 1; |
196 | high = ((u64)slow_virt_to_phys(src) >> 32); | |
19b6a85b | 197 | ret = native_write_msr_safe(msr_kvm_system_time, low, high); |
f6e16d5a GH |
198 | printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n", |
199 | cpu, high, low, txt); | |
838815a7 | 200 | |
19b6a85b | 201 | return ret; |
790c73f6 GOC |
202 | } |
203 | ||
b74f05d6 MT |
204 | static void kvm_save_sched_clock_state(void) |
205 | { | |
206 | } | |
207 | ||
208 | static void kvm_restore_sched_clock_state(void) | |
209 | { | |
210 | kvm_register_clock("primary cpu clock, resume"); | |
211 | } | |
212 | ||
b8ba5f10 | 213 | #ifdef CONFIG_X86_LOCAL_APIC |
148f9bb8 | 214 | static void kvm_setup_secondary_clock(void) |
790c73f6 GOC |
215 | { |
216 | /* | |
217 | * Now that the first cpu already had this clocksource initialized, | |
218 | * we shouldn't fail. | |
219 | */ | |
f6e16d5a | 220 | WARN_ON(kvm_register_clock("secondary cpu clock")); |
790c73f6 | 221 | } |
b8ba5f10 | 222 | #endif |
790c73f6 | 223 | |
1e977aa1 GC |
224 | /* |
225 | * After the clock is registered, the host will keep writing to the | |
226 | * registered memory location. If the guest happens to shutdown, this memory | |
227 | * won't be valid. In cases like kexec, in which you install a new kernel, this | |
228 | * means a random memory location will be kept being written. So before any | |
6a6256f9 | 229 | * kind of shutdown from our side, we unregister the clock by writing anything |
1e977aa1 GC |
230 | * that does not have the 'enable' bit set in the msr |
231 | */ | |
2965faa5 | 232 | #ifdef CONFIG_KEXEC_CORE |
1e977aa1 GC |
233 | static void kvm_crash_shutdown(struct pt_regs *regs) |
234 | { | |
838815a7 | 235 | native_write_msr(msr_kvm_system_time, 0, 0); |
d910f5c1 | 236 | kvm_disable_steal_time(); |
1e977aa1 GC |
237 | native_machine_crash_shutdown(regs); |
238 | } | |
239 | #endif | |
240 | ||
241 | static void kvm_shutdown(void) | |
242 | { | |
838815a7 | 243 | native_write_msr(msr_kvm_system_time, 0, 0); |
d910f5c1 | 244 | kvm_disable_steal_time(); |
1e977aa1 GC |
245 | native_machine_shutdown(); |
246 | } | |
247 | ||
790c73f6 GOC |
248 | void __init kvmclock_init(void) |
249 | { | |
0ad83caa | 250 | struct pvclock_vcpu_time_info *vcpu_time; |
7069ed67 | 251 | unsigned long mem; |
0ad83caa LC |
252 | int size, cpu; |
253 | u8 flags; | |
ed55705d MT |
254 | |
255 | size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS); | |
7069ed67 | 256 | |
790c73f6 GOC |
257 | if (!kvm_para_available()) |
258 | return; | |
259 | ||
838815a7 GC |
260 | if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) { |
261 | msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW; | |
262 | msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW; | |
263 | } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE))) | |
264 | return; | |
265 | ||
266 | printk(KERN_INFO "kvm-clock: Using msrs %x and %x", | |
267 | msr_kvm_system_time, msr_kvm_wall_clock); | |
268 | ||
ed55705d | 269 | mem = memblock_alloc(size, PAGE_SIZE); |
7069ed67 MT |
270 | if (!mem) |
271 | return; | |
272 | hv_clock = __va(mem); | |
07868fc6 | 273 | memset(hv_clock, 0, size); |
7069ed67 | 274 | |
0d75de4a | 275 | if (kvm_register_clock("primary cpu clock")) { |
7069ed67 | 276 | hv_clock = NULL; |
ed55705d | 277 | memblock_free(mem, size); |
838815a7 | 278 | return; |
7069ed67 | 279 | } |
72c930dc RK |
280 | |
281 | if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT)) | |
282 | pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT); | |
283 | ||
284 | cpu = get_cpu(); | |
285 | vcpu_time = &hv_clock[cpu].pvti; | |
286 | flags = pvclock_read_flags(vcpu_time); | |
287 | ||
288 | kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT); | |
289 | put_cpu(); | |
290 | ||
838815a7 | 291 | x86_platform.calibrate_tsc = kvm_get_tsc_khz; |
a4497a86 | 292 | x86_platform.calibrate_cpu = kvm_get_tsc_khz; |
838815a7 GC |
293 | x86_platform.get_wallclock = kvm_get_wallclock; |
294 | x86_platform.set_wallclock = kvm_set_wallclock; | |
b8ba5f10 | 295 | #ifdef CONFIG_X86_LOCAL_APIC |
df156f90 | 296 | x86_cpuinit.early_percpu_clock_init = |
838815a7 | 297 | kvm_setup_secondary_clock; |
b8ba5f10 | 298 | #endif |
b74f05d6 MT |
299 | x86_platform.save_sched_clock_state = kvm_save_sched_clock_state; |
300 | x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state; | |
838815a7 | 301 | machine_ops.shutdown = kvm_shutdown; |
2965faa5 | 302 | #ifdef CONFIG_KEXEC_CORE |
838815a7 | 303 | machine_ops.crash_shutdown = kvm_crash_shutdown; |
1e977aa1 | 304 | #endif |
838815a7 | 305 | kvm_get_preset_lpj(); |
b01cc1b0 | 306 | clocksource_register_hz(&kvm_clock, NSEC_PER_SEC); |
838815a7 | 307 | pv_info.name = "KVM"; |
790c73f6 | 308 | } |
3dc4f7cf MT |
309 | |
310 | int __init kvm_setup_vsyscall_timeinfo(void) | |
311 | { | |
312 | #ifdef CONFIG_X86_64 | |
313 | int cpu; | |
3dc4f7cf MT |
314 | u8 flags; |
315 | struct pvclock_vcpu_time_info *vcpu_time; | |
316 | unsigned int size; | |
317 | ||
fe1140cc JK |
318 | if (!hv_clock) |
319 | return 0; | |
320 | ||
ed55705d | 321 | size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS); |
3dc4f7cf | 322 | |
c6338ce4 | 323 | cpu = get_cpu(); |
3dc4f7cf MT |
324 | |
325 | vcpu_time = &hv_clock[cpu].pvti; | |
326 | flags = pvclock_read_flags(vcpu_time); | |
327 | ||
328 | if (!(flags & PVCLOCK_TSC_STABLE_BIT)) { | |
c6338ce4 | 329 | put_cpu(); |
3dc4f7cf MT |
330 | return 1; |
331 | } | |
332 | ||
c6338ce4 | 333 | put_cpu(); |
3dc4f7cf MT |
334 | |
335 | kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK; | |
336 | #endif | |
337 | return 0; | |
338 | } |