]>
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> |
736decac TG |
27 | |
28 | #include <asm/x86_init.h> | |
1e977aa1 | 29 | #include <asm/reboot.h> |
790c73f6 | 30 | |
790c73f6 | 31 | static int kvmclock = 1; |
838815a7 GC |
32 | static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME; |
33 | static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK; | |
790c73f6 GOC |
34 | |
35 | static int parse_no_kvmclock(char *arg) | |
36 | { | |
37 | kvmclock = 0; | |
38 | return 0; | |
39 | } | |
40 | early_param("no-kvmclock", parse_no_kvmclock); | |
41 | ||
42 | /* The hypervisor will put information about time periodically here */ | |
3dc4f7cf | 43 | static struct pvclock_vsyscall_time_info *hv_clock; |
f6e16d5a | 44 | static struct pvclock_wall_clock wall_clock; |
790c73f6 | 45 | |
790c73f6 GOC |
46 | /* |
47 | * The wallclock is the time of day when we booted. Since then, some time may | |
48 | * have elapsed since the hypervisor wrote the data. So we try to account for | |
49 | * that with system time | |
50 | */ | |
2ddfd20e | 51 | static unsigned long kvm_get_wallclock(void) |
790c73f6 | 52 | { |
f6e16d5a | 53 | struct pvclock_vcpu_time_info *vcpu_time; |
790c73f6 | 54 | struct timespec ts; |
790c73f6 | 55 | int low, high; |
7069ed67 | 56 | int cpu; |
790c73f6 | 57 | |
a20316d2 GC |
58 | low = (int)__pa_symbol(&wall_clock); |
59 | high = ((u64)__pa_symbol(&wall_clock) >> 32); | |
838815a7 GC |
60 | |
61 | native_write_msr(msr_kvm_wall_clock, low, high); | |
790c73f6 | 62 | |
7069ed67 MT |
63 | preempt_disable(); |
64 | cpu = smp_processor_id(); | |
65 | ||
3dc4f7cf | 66 | vcpu_time = &hv_clock[cpu].pvti; |
f6e16d5a | 67 | pvclock_read_wallclock(&wall_clock, vcpu_time, &ts); |
7069ed67 MT |
68 | |
69 | preempt_enable(); | |
790c73f6 | 70 | |
f6e16d5a | 71 | return ts.tv_sec; |
790c73f6 GOC |
72 | } |
73 | ||
2ddfd20e | 74 | static int kvm_set_wallclock(unsigned long now) |
790c73f6 | 75 | { |
f6e16d5a | 76 | return -1; |
790c73f6 GOC |
77 | } |
78 | ||
790c73f6 GOC |
79 | static cycle_t kvm_clock_read(void) |
80 | { | |
f6e16d5a GH |
81 | struct pvclock_vcpu_time_info *src; |
82 | cycle_t ret; | |
7069ed67 | 83 | int cpu; |
790c73f6 | 84 | |
95ef1e52 | 85 | preempt_disable_notrace(); |
7069ed67 | 86 | cpu = smp_processor_id(); |
3dc4f7cf | 87 | src = &hv_clock[cpu].pvti; |
f6e16d5a | 88 | ret = pvclock_clocksource_read(src); |
95ef1e52 | 89 | preempt_enable_notrace(); |
f6e16d5a | 90 | return ret; |
790c73f6 | 91 | } |
f6e16d5a | 92 | |
8e19608e MD |
93 | static cycle_t kvm_clock_get_cycles(struct clocksource *cs) |
94 | { | |
95 | return kvm_clock_read(); | |
96 | } | |
97 | ||
0293615f GC |
98 | /* |
99 | * If we don't do that, there is the possibility that the guest | |
100 | * will calibrate under heavy load - thus, getting a lower lpj - | |
101 | * and execute the delays themselves without load. This is wrong, | |
102 | * because no delay loop can finish beforehand. | |
103 | * Any heuristics is subject to fail, because ultimately, a large | |
104 | * poll of guests can be running and trouble each other. So we preset | |
105 | * lpj here | |
106 | */ | |
107 | static unsigned long kvm_get_tsc_khz(void) | |
108 | { | |
e93353c9 | 109 | struct pvclock_vcpu_time_info *src; |
7069ed67 MT |
110 | int cpu; |
111 | unsigned long tsc_khz; | |
112 | ||
113 | preempt_disable(); | |
114 | cpu = smp_processor_id(); | |
3dc4f7cf | 115 | src = &hv_clock[cpu].pvti; |
7069ed67 MT |
116 | tsc_khz = pvclock_tsc_khz(src); |
117 | preempt_enable(); | |
118 | return tsc_khz; | |
0293615f GC |
119 | } |
120 | ||
121 | static void kvm_get_preset_lpj(void) | |
122 | { | |
0293615f GC |
123 | unsigned long khz; |
124 | u64 lpj; | |
125 | ||
e93353c9 | 126 | khz = kvm_get_tsc_khz(); |
0293615f GC |
127 | |
128 | lpj = ((u64)khz * 1000); | |
129 | do_div(lpj, HZ); | |
130 | preset_lpj = lpj; | |
131 | } | |
132 | ||
3b5d56b9 EM |
133 | bool kvm_check_and_clear_guest_paused(void) |
134 | { | |
135 | bool ret = false; | |
136 | struct pvclock_vcpu_time_info *src; | |
7069ed67 MT |
137 | int cpu = smp_processor_id(); |
138 | ||
139 | if (!hv_clock) | |
140 | return ret; | |
3b5d56b9 | 141 | |
3dc4f7cf | 142 | src = &hv_clock[cpu].pvti; |
3b5d56b9 | 143 | if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) { |
7069ed67 | 144 | src->flags &= ~PVCLOCK_GUEST_STOPPED; |
3b5d56b9 EM |
145 | ret = true; |
146 | } | |
147 | ||
148 | return ret; | |
149 | } | |
3b5d56b9 | 150 | |
790c73f6 GOC |
151 | static struct clocksource kvm_clock = { |
152 | .name = "kvm-clock", | |
8e19608e | 153 | .read = kvm_clock_get_cycles, |
790c73f6 GOC |
154 | .rating = 400, |
155 | .mask = CLOCKSOURCE_MASK(64), | |
790c73f6 GOC |
156 | .flags = CLOCK_SOURCE_IS_CONTINUOUS, |
157 | }; | |
158 | ||
ca3f1017 | 159 | int kvm_register_clock(char *txt) |
790c73f6 GOC |
160 | { |
161 | int cpu = smp_processor_id(); | |
19b6a85b | 162 | int low, high, ret; |
3dc4f7cf | 163 | struct pvclock_vcpu_time_info *src = &hv_clock[cpu].pvti; |
19b6a85b | 164 | |
7069ed67 MT |
165 | low = (int)__pa(src) | 1; |
166 | high = ((u64)__pa(src) >> 32); | |
19b6a85b | 167 | ret = native_write_msr_safe(msr_kvm_system_time, low, high); |
f6e16d5a GH |
168 | printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n", |
169 | cpu, high, low, txt); | |
838815a7 | 170 | |
19b6a85b | 171 | return ret; |
790c73f6 GOC |
172 | } |
173 | ||
b74f05d6 MT |
174 | static void kvm_save_sched_clock_state(void) |
175 | { | |
176 | } | |
177 | ||
178 | static void kvm_restore_sched_clock_state(void) | |
179 | { | |
180 | kvm_register_clock("primary cpu clock, resume"); | |
181 | } | |
182 | ||
b8ba5f10 | 183 | #ifdef CONFIG_X86_LOCAL_APIC |
23a14b9e | 184 | static void __cpuinit kvm_setup_secondary_clock(void) |
790c73f6 GOC |
185 | { |
186 | /* | |
187 | * Now that the first cpu already had this clocksource initialized, | |
188 | * we shouldn't fail. | |
189 | */ | |
f6e16d5a | 190 | WARN_ON(kvm_register_clock("secondary cpu clock")); |
790c73f6 | 191 | } |
b8ba5f10 | 192 | #endif |
790c73f6 | 193 | |
1e977aa1 GC |
194 | /* |
195 | * After the clock is registered, the host will keep writing to the | |
196 | * registered memory location. If the guest happens to shutdown, this memory | |
197 | * won't be valid. In cases like kexec, in which you install a new kernel, this | |
198 | * means a random memory location will be kept being written. So before any | |
199 | * kind of shutdown from our side, we unregister the clock by writting anything | |
200 | * that does not have the 'enable' bit set in the msr | |
201 | */ | |
202 | #ifdef CONFIG_KEXEC | |
203 | static void kvm_crash_shutdown(struct pt_regs *regs) | |
204 | { | |
838815a7 | 205 | native_write_msr(msr_kvm_system_time, 0, 0); |
d910f5c1 | 206 | kvm_disable_steal_time(); |
1e977aa1 GC |
207 | native_machine_crash_shutdown(regs); |
208 | } | |
209 | #endif | |
210 | ||
211 | static void kvm_shutdown(void) | |
212 | { | |
838815a7 | 213 | native_write_msr(msr_kvm_system_time, 0, 0); |
d910f5c1 | 214 | kvm_disable_steal_time(); |
1e977aa1 GC |
215 | native_machine_shutdown(); |
216 | } | |
217 | ||
790c73f6 GOC |
218 | void __init kvmclock_init(void) |
219 | { | |
7069ed67 MT |
220 | unsigned long mem; |
221 | ||
790c73f6 GOC |
222 | if (!kvm_para_available()) |
223 | return; | |
224 | ||
838815a7 GC |
225 | if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) { |
226 | msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW; | |
227 | msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW; | |
228 | } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE))) | |
229 | return; | |
230 | ||
231 | printk(KERN_INFO "kvm-clock: Using msrs %x and %x", | |
232 | msr_kvm_system_time, msr_kvm_wall_clock); | |
233 | ||
3dc4f7cf | 234 | mem = memblock_alloc(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS, |
7069ed67 MT |
235 | PAGE_SIZE); |
236 | if (!mem) | |
237 | return; | |
238 | hv_clock = __va(mem); | |
239 | ||
240 | if (kvm_register_clock("boot clock")) { | |
241 | hv_clock = NULL; | |
242 | memblock_free(mem, | |
3dc4f7cf | 243 | sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS); |
838815a7 | 244 | return; |
7069ed67 | 245 | } |
838815a7 GC |
246 | pv_time_ops.sched_clock = kvm_clock_read; |
247 | x86_platform.calibrate_tsc = kvm_get_tsc_khz; | |
248 | x86_platform.get_wallclock = kvm_get_wallclock; | |
249 | x86_platform.set_wallclock = kvm_set_wallclock; | |
b8ba5f10 | 250 | #ifdef CONFIG_X86_LOCAL_APIC |
df156f90 | 251 | x86_cpuinit.early_percpu_clock_init = |
838815a7 | 252 | kvm_setup_secondary_clock; |
b8ba5f10 | 253 | #endif |
b74f05d6 MT |
254 | x86_platform.save_sched_clock_state = kvm_save_sched_clock_state; |
255 | x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state; | |
838815a7 | 256 | machine_ops.shutdown = kvm_shutdown; |
1e977aa1 | 257 | #ifdef CONFIG_KEXEC |
838815a7 | 258 | machine_ops.crash_shutdown = kvm_crash_shutdown; |
1e977aa1 | 259 | #endif |
838815a7 | 260 | kvm_get_preset_lpj(); |
b01cc1b0 | 261 | clocksource_register_hz(&kvm_clock, NSEC_PER_SEC); |
838815a7 GC |
262 | pv_info.paravirt_enabled = 1; |
263 | pv_info.name = "KVM"; | |
3a0d7256 GC |
264 | |
265 | if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT)) | |
266 | pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT); | |
790c73f6 | 267 | } |
3dc4f7cf MT |
268 | |
269 | int __init kvm_setup_vsyscall_timeinfo(void) | |
270 | { | |
271 | #ifdef CONFIG_X86_64 | |
272 | int cpu; | |
273 | int ret; | |
274 | u8 flags; | |
275 | struct pvclock_vcpu_time_info *vcpu_time; | |
276 | unsigned int size; | |
277 | ||
278 | size = sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS; | |
279 | ||
280 | preempt_disable(); | |
281 | cpu = smp_processor_id(); | |
282 | ||
283 | vcpu_time = &hv_clock[cpu].pvti; | |
284 | flags = pvclock_read_flags(vcpu_time); | |
285 | ||
286 | if (!(flags & PVCLOCK_TSC_STABLE_BIT)) { | |
287 | preempt_enable(); | |
288 | return 1; | |
289 | } | |
290 | ||
291 | if ((ret = pvclock_init_vsyscall(hv_clock, size))) { | |
292 | preempt_enable(); | |
293 | return ret; | |
294 | } | |
295 | ||
296 | preempt_enable(); | |
297 | ||
298 | kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK; | |
299 | #endif | |
300 | return 0; | |
301 | } |