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