]>
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> |
e6017571 | 28 | #include <linux/sched/clock.h> |
736decac TG |
29 | |
30 | #include <asm/x86_init.h> | |
1e977aa1 | 31 | #include <asm/reboot.h> |
f4066c2b | 32 | #include <asm/kvmclock.h> |
790c73f6 | 33 | |
404f6aac | 34 | static int kvmclock __ro_after_init = 1; |
838815a7 GC |
35 | static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME; |
36 | static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK; | |
a5a1d1c2 | 37 | static u64 kvm_sched_clock_offset; |
790c73f6 GOC |
38 | |
39 | static int parse_no_kvmclock(char *arg) | |
40 | { | |
41 | kvmclock = 0; | |
42 | return 0; | |
43 | } | |
44 | early_param("no-kvmclock", parse_no_kvmclock); | |
45 | ||
46 | /* The hypervisor will put information about time periodically here */ | |
3dc4f7cf | 47 | static struct pvclock_vsyscall_time_info *hv_clock; |
f6e16d5a | 48 | static struct pvclock_wall_clock wall_clock; |
790c73f6 | 49 | |
dac16fba AL |
50 | struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void) |
51 | { | |
52 | return hv_clock; | |
53 | } | |
f4066c2b | 54 | EXPORT_SYMBOL_GPL(pvclock_pvti_cpu0_va); |
dac16fba | 55 | |
790c73f6 GOC |
56 | /* |
57 | * The wallclock is the time of day when we booted. Since then, some time may | |
58 | * have elapsed since the hypervisor wrote the data. So we try to account for | |
59 | * that with system time | |
60 | */ | |
3565184e | 61 | static void kvm_get_wallclock(struct timespec *now) |
790c73f6 | 62 | { |
f6e16d5a | 63 | struct pvclock_vcpu_time_info *vcpu_time; |
790c73f6 | 64 | int low, high; |
7069ed67 | 65 | int cpu; |
790c73f6 | 66 | |
a20316d2 GC |
67 | low = (int)__pa_symbol(&wall_clock); |
68 | high = ((u64)__pa_symbol(&wall_clock) >> 32); | |
838815a7 GC |
69 | |
70 | native_write_msr(msr_kvm_wall_clock, low, high); | |
790c73f6 | 71 | |
c6338ce4 | 72 | cpu = get_cpu(); |
7069ed67 | 73 | |
3dc4f7cf | 74 | vcpu_time = &hv_clock[cpu].pvti; |
3565184e | 75 | pvclock_read_wallclock(&wall_clock, vcpu_time, now); |
7069ed67 | 76 | |
c6338ce4 | 77 | put_cpu(); |
790c73f6 GOC |
78 | } |
79 | ||
3565184e | 80 | static int kvm_set_wallclock(const struct timespec *now) |
790c73f6 | 81 | { |
f6e16d5a | 82 | return -1; |
790c73f6 GOC |
83 | } |
84 | ||
a5a1d1c2 | 85 | static u64 kvm_clock_read(void) |
790c73f6 | 86 | { |
f6e16d5a | 87 | struct pvclock_vcpu_time_info *src; |
a5a1d1c2 | 88 | u64 ret; |
7069ed67 | 89 | int cpu; |
790c73f6 | 90 | |
95ef1e52 | 91 | preempt_disable_notrace(); |
7069ed67 | 92 | cpu = smp_processor_id(); |
3dc4f7cf | 93 | src = &hv_clock[cpu].pvti; |
f6e16d5a | 94 | ret = pvclock_clocksource_read(src); |
95ef1e52 | 95 | preempt_enable_notrace(); |
f6e16d5a | 96 | return ret; |
790c73f6 | 97 | } |
f6e16d5a | 98 | |
a5a1d1c2 | 99 | static u64 kvm_clock_get_cycles(struct clocksource *cs) |
8e19608e MD |
100 | { |
101 | return kvm_clock_read(); | |
102 | } | |
103 | ||
a5a1d1c2 | 104 | static u64 kvm_sched_clock_read(void) |
72c930dc RK |
105 | { |
106 | return kvm_clock_read() - kvm_sched_clock_offset; | |
107 | } | |
108 | ||
109 | static inline void kvm_sched_clock_init(bool stable) | |
110 | { | |
111 | if (!stable) { | |
112 | pv_time_ops.sched_clock = kvm_clock_read; | |
acb04058 | 113 | clear_sched_clock_stable(); |
72c930dc RK |
114 | return; |
115 | } | |
116 | ||
117 | kvm_sched_clock_offset = kvm_clock_read(); | |
118 | pv_time_ops.sched_clock = kvm_sched_clock_read; | |
72c930dc RK |
119 | |
120 | printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n", | |
121 | kvm_sched_clock_offset); | |
122 | ||
123 | BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) > | |
124 | sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time)); | |
125 | } | |
126 | ||
0293615f GC |
127 | /* |
128 | * If we don't do that, there is the possibility that the guest | |
129 | * will calibrate under heavy load - thus, getting a lower lpj - | |
130 | * and execute the delays themselves without load. This is wrong, | |
131 | * because no delay loop can finish beforehand. | |
132 | * Any heuristics is subject to fail, because ultimately, a large | |
133 | * poll of guests can be running and trouble each other. So we preset | |
134 | * lpj here | |
135 | */ | |
136 | static unsigned long kvm_get_tsc_khz(void) | |
137 | { | |
e93353c9 | 138 | struct pvclock_vcpu_time_info *src; |
7069ed67 MT |
139 | int cpu; |
140 | unsigned long tsc_khz; | |
141 | ||
c6338ce4 | 142 | cpu = get_cpu(); |
3dc4f7cf | 143 | src = &hv_clock[cpu].pvti; |
7069ed67 | 144 | tsc_khz = pvclock_tsc_khz(src); |
c6338ce4 | 145 | put_cpu(); |
7069ed67 | 146 | return tsc_khz; |
0293615f GC |
147 | } |
148 | ||
149 | static void kvm_get_preset_lpj(void) | |
150 | { | |
0293615f GC |
151 | unsigned long khz; |
152 | u64 lpj; | |
153 | ||
e93353c9 | 154 | khz = kvm_get_tsc_khz(); |
0293615f GC |
155 | |
156 | lpj = ((u64)khz * 1000); | |
157 | do_div(lpj, HZ); | |
158 | preset_lpj = lpj; | |
159 | } | |
160 | ||
3b5d56b9 EM |
161 | bool kvm_check_and_clear_guest_paused(void) |
162 | { | |
163 | bool ret = false; | |
164 | struct pvclock_vcpu_time_info *src; | |
7069ed67 MT |
165 | int cpu = smp_processor_id(); |
166 | ||
167 | if (!hv_clock) | |
168 | return ret; | |
3b5d56b9 | 169 | |
3dc4f7cf | 170 | src = &hv_clock[cpu].pvti; |
3b5d56b9 | 171 | if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) { |
7069ed67 | 172 | src->flags &= ~PVCLOCK_GUEST_STOPPED; |
d63285e9 | 173 | pvclock_touch_watchdogs(); |
3b5d56b9 EM |
174 | ret = true; |
175 | } | |
176 | ||
177 | return ret; | |
178 | } | |
3b5d56b9 | 179 | |
f4066c2b | 180 | struct clocksource kvm_clock = { |
790c73f6 | 181 | .name = "kvm-clock", |
8e19608e | 182 | .read = kvm_clock_get_cycles, |
790c73f6 GOC |
183 | .rating = 400, |
184 | .mask = CLOCKSOURCE_MASK(64), | |
790c73f6 GOC |
185 | .flags = CLOCK_SOURCE_IS_CONTINUOUS, |
186 | }; | |
f4066c2b | 187 | EXPORT_SYMBOL_GPL(kvm_clock); |
790c73f6 | 188 | |
ca3f1017 | 189 | int kvm_register_clock(char *txt) |
790c73f6 GOC |
190 | { |
191 | int cpu = smp_processor_id(); | |
19b6a85b | 192 | int low, high, ret; |
fe1140cc JK |
193 | struct pvclock_vcpu_time_info *src; |
194 | ||
195 | if (!hv_clock) | |
196 | return 0; | |
19b6a85b | 197 | |
fe1140cc | 198 | src = &hv_clock[cpu].pvti; |
5dfd486c DH |
199 | low = (int)slow_virt_to_phys(src) | 1; |
200 | high = ((u64)slow_virt_to_phys(src) >> 32); | |
19b6a85b | 201 | ret = native_write_msr_safe(msr_kvm_system_time, low, high); |
f6e16d5a GH |
202 | printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n", |
203 | cpu, high, low, txt); | |
838815a7 | 204 | |
19b6a85b | 205 | return ret; |
790c73f6 GOC |
206 | } |
207 | ||
b74f05d6 MT |
208 | static void kvm_save_sched_clock_state(void) |
209 | { | |
210 | } | |
211 | ||
212 | static void kvm_restore_sched_clock_state(void) | |
213 | { | |
214 | kvm_register_clock("primary cpu clock, resume"); | |
215 | } | |
216 | ||
b8ba5f10 | 217 | #ifdef CONFIG_X86_LOCAL_APIC |
148f9bb8 | 218 | static void kvm_setup_secondary_clock(void) |
790c73f6 GOC |
219 | { |
220 | /* | |
221 | * Now that the first cpu already had this clocksource initialized, | |
222 | * we shouldn't fail. | |
223 | */ | |
f6e16d5a | 224 | WARN_ON(kvm_register_clock("secondary cpu clock")); |
790c73f6 | 225 | } |
b8ba5f10 | 226 | #endif |
790c73f6 | 227 | |
1e977aa1 GC |
228 | /* |
229 | * After the clock is registered, the host will keep writing to the | |
230 | * registered memory location. If the guest happens to shutdown, this memory | |
231 | * won't be valid. In cases like kexec, in which you install a new kernel, this | |
232 | * means a random memory location will be kept being written. So before any | |
6a6256f9 | 233 | * kind of shutdown from our side, we unregister the clock by writing anything |
1e977aa1 GC |
234 | * that does not have the 'enable' bit set in the msr |
235 | */ | |
2965faa5 | 236 | #ifdef CONFIG_KEXEC_CORE |
1e977aa1 GC |
237 | static void kvm_crash_shutdown(struct pt_regs *regs) |
238 | { | |
838815a7 | 239 | native_write_msr(msr_kvm_system_time, 0, 0); |
d910f5c1 | 240 | kvm_disable_steal_time(); |
1e977aa1 GC |
241 | native_machine_crash_shutdown(regs); |
242 | } | |
243 | #endif | |
244 | ||
245 | static void kvm_shutdown(void) | |
246 | { | |
838815a7 | 247 | native_write_msr(msr_kvm_system_time, 0, 0); |
d910f5c1 | 248 | kvm_disable_steal_time(); |
1e977aa1 GC |
249 | native_machine_shutdown(); |
250 | } | |
251 | ||
790c73f6 GOC |
252 | void __init kvmclock_init(void) |
253 | { | |
0ad83caa | 254 | struct pvclock_vcpu_time_info *vcpu_time; |
7069ed67 | 255 | unsigned long mem; |
0ad83caa LC |
256 | int size, cpu; |
257 | u8 flags; | |
ed55705d MT |
258 | |
259 | size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS); | |
7069ed67 | 260 | |
790c73f6 GOC |
261 | if (!kvm_para_available()) |
262 | return; | |
263 | ||
838815a7 GC |
264 | if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) { |
265 | msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW; | |
266 | msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW; | |
267 | } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE))) | |
268 | return; | |
269 | ||
270 | printk(KERN_INFO "kvm-clock: Using msrs %x and %x", | |
271 | msr_kvm_system_time, msr_kvm_wall_clock); | |
272 | ||
ed55705d | 273 | mem = memblock_alloc(size, PAGE_SIZE); |
7069ed67 MT |
274 | if (!mem) |
275 | return; | |
276 | hv_clock = __va(mem); | |
07868fc6 | 277 | memset(hv_clock, 0, size); |
7069ed67 | 278 | |
0d75de4a | 279 | if (kvm_register_clock("primary cpu clock")) { |
7069ed67 | 280 | hv_clock = NULL; |
ed55705d | 281 | memblock_free(mem, size); |
838815a7 | 282 | return; |
7069ed67 | 283 | } |
72c930dc RK |
284 | |
285 | if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT)) | |
286 | pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT); | |
287 | ||
288 | cpu = get_cpu(); | |
289 | vcpu_time = &hv_clock[cpu].pvti; | |
290 | flags = pvclock_read_flags(vcpu_time); | |
291 | ||
292 | kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT); | |
293 | put_cpu(); | |
294 | ||
838815a7 | 295 | x86_platform.calibrate_tsc = kvm_get_tsc_khz; |
a4497a86 | 296 | x86_platform.calibrate_cpu = kvm_get_tsc_khz; |
838815a7 GC |
297 | x86_platform.get_wallclock = kvm_get_wallclock; |
298 | x86_platform.set_wallclock = kvm_set_wallclock; | |
b8ba5f10 | 299 | #ifdef CONFIG_X86_LOCAL_APIC |
df156f90 | 300 | x86_cpuinit.early_percpu_clock_init = |
838815a7 | 301 | kvm_setup_secondary_clock; |
b8ba5f10 | 302 | #endif |
b74f05d6 MT |
303 | x86_platform.save_sched_clock_state = kvm_save_sched_clock_state; |
304 | x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state; | |
838815a7 | 305 | machine_ops.shutdown = kvm_shutdown; |
2965faa5 | 306 | #ifdef CONFIG_KEXEC_CORE |
838815a7 | 307 | machine_ops.crash_shutdown = kvm_crash_shutdown; |
1e977aa1 | 308 | #endif |
838815a7 | 309 | kvm_get_preset_lpj(); |
b01cc1b0 | 310 | clocksource_register_hz(&kvm_clock, NSEC_PER_SEC); |
838815a7 | 311 | pv_info.name = "KVM"; |
790c73f6 | 312 | } |
3dc4f7cf MT |
313 | |
314 | int __init kvm_setup_vsyscall_timeinfo(void) | |
315 | { | |
316 | #ifdef CONFIG_X86_64 | |
317 | int cpu; | |
3dc4f7cf MT |
318 | u8 flags; |
319 | struct pvclock_vcpu_time_info *vcpu_time; | |
320 | unsigned int size; | |
321 | ||
fe1140cc JK |
322 | if (!hv_clock) |
323 | return 0; | |
324 | ||
ed55705d | 325 | size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS); |
3dc4f7cf | 326 | |
c6338ce4 | 327 | cpu = get_cpu(); |
3dc4f7cf MT |
328 | |
329 | vcpu_time = &hv_clock[cpu].pvti; | |
330 | flags = pvclock_read_flags(vcpu_time); | |
331 | ||
332 | if (!(flags & PVCLOCK_TSC_STABLE_BIT)) { | |
c6338ce4 | 333 | put_cpu(); |
3dc4f7cf MT |
334 | return 1; |
335 | } | |
336 | ||
c6338ce4 | 337 | put_cpu(); |
3dc4f7cf MT |
338 | |
339 | kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK; | |
340 | #endif | |
341 | return 0; | |
342 | } |