1 /* KVM paravirtual clock driver. A clocksource implementation
2 Copyright (C) 2008 Glauber de Oliveira Costa, Red Hat Inc.
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.
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.
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
19 #include <linux/clocksource.h>
20 #include <linux/kvm_para.h>
21 #include <asm/pvclock.h>
24 #include <linux/percpu.h>
25 #include <linux/hardirq.h>
26 #include <linux/memblock.h>
27 #include <linux/sched.h>
28 #include <linux/sched/clock.h>
30 #include <asm/x86_init.h>
31 #include <asm/reboot.h>
32 #include <asm/kvmclock.h>
34 static int kvmclock __ro_after_init = 1;
35 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
36 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
37 static u64 kvm_sched_clock_offset;
39 static int parse_no_kvmclock(char *arg)
44 early_param("no-kvmclock", parse_no_kvmclock);
46 /* The hypervisor will put information about time periodically here */
47 static struct pvclock_vsyscall_time_info *hv_clock;
48 static struct pvclock_wall_clock wall_clock;
50 struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void)
54 EXPORT_SYMBOL_GPL(pvclock_pvti_cpu0_va);
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
61 static void kvm_get_wallclock(struct timespec *now)
63 struct pvclock_vcpu_time_info *vcpu_time;
67 low = (int)__pa_symbol(&wall_clock);
68 high = ((u64)__pa_symbol(&wall_clock) >> 32);
70 native_write_msr(msr_kvm_wall_clock, low, high);
74 vcpu_time = &hv_clock[cpu].pvti;
75 pvclock_read_wallclock(&wall_clock, vcpu_time, now);
80 static int kvm_set_wallclock(const struct timespec *now)
85 static u64 kvm_clock_read(void)
87 struct pvclock_vcpu_time_info *src;
91 preempt_disable_notrace();
92 cpu = smp_processor_id();
93 src = &hv_clock[cpu].pvti;
94 ret = pvclock_clocksource_read(src);
95 preempt_enable_notrace();
99 static u64 kvm_clock_get_cycles(struct clocksource *cs)
101 return kvm_clock_read();
104 static u64 kvm_sched_clock_read(void)
106 return kvm_clock_read() - kvm_sched_clock_offset;
109 static inline void kvm_sched_clock_init(bool stable)
112 pv_time_ops.sched_clock = kvm_clock_read;
113 clear_sched_clock_stable();
117 kvm_sched_clock_offset = kvm_clock_read();
118 pv_time_ops.sched_clock = kvm_sched_clock_read;
120 printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
121 kvm_sched_clock_offset);
123 BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
124 sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
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
136 static unsigned long kvm_get_tsc_khz(void)
138 struct pvclock_vcpu_time_info *src;
140 unsigned long tsc_khz;
143 src = &hv_clock[cpu].pvti;
144 tsc_khz = pvclock_tsc_khz(src);
149 static void kvm_get_preset_lpj(void)
154 khz = kvm_get_tsc_khz();
156 lpj = ((u64)khz * 1000);
161 bool kvm_check_and_clear_guest_paused(void)
164 struct pvclock_vcpu_time_info *src;
165 int cpu = smp_processor_id();
170 src = &hv_clock[cpu].pvti;
171 if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
172 src->flags &= ~PVCLOCK_GUEST_STOPPED;
173 pvclock_touch_watchdogs();
180 struct clocksource kvm_clock = {
182 .read = kvm_clock_get_cycles,
184 .mask = CLOCKSOURCE_MASK(64),
185 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
187 EXPORT_SYMBOL_GPL(kvm_clock);
189 int kvm_register_clock(char *txt)
191 int cpu = smp_processor_id();
193 struct pvclock_vcpu_time_info *src;
198 src = &hv_clock[cpu].pvti;
199 low = (int)slow_virt_to_phys(src) | 1;
200 high = ((u64)slow_virt_to_phys(src) >> 32);
201 ret = native_write_msr_safe(msr_kvm_system_time, low, high);
202 printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
203 cpu, high, low, txt);
208 static void kvm_save_sched_clock_state(void)
212 static void kvm_restore_sched_clock_state(void)
214 kvm_register_clock("primary cpu clock, resume");
217 #ifdef CONFIG_X86_LOCAL_APIC
218 static void kvm_setup_secondary_clock(void)
221 * Now that the first cpu already had this clocksource initialized,
224 WARN_ON(kvm_register_clock("secondary cpu clock"));
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
233 * kind of shutdown from our side, we unregister the clock by writing anything
234 * that does not have the 'enable' bit set in the msr
236 #ifdef CONFIG_KEXEC_CORE
237 static void kvm_crash_shutdown(struct pt_regs *regs)
239 native_write_msr(msr_kvm_system_time, 0, 0);
240 kvm_disable_steal_time();
241 native_machine_crash_shutdown(regs);
245 static void kvm_shutdown(void)
247 native_write_msr(msr_kvm_system_time, 0, 0);
248 kvm_disable_steal_time();
249 native_machine_shutdown();
252 void __init kvmclock_init(void)
254 struct pvclock_vcpu_time_info *vcpu_time;
259 size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
261 if (!kvm_para_available())
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)))
270 printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
271 msr_kvm_system_time, msr_kvm_wall_clock);
273 mem = memblock_alloc(size, PAGE_SIZE);
276 hv_clock = __va(mem);
277 memset(hv_clock, 0, size);
279 if (kvm_register_clock("primary cpu clock")) {
281 memblock_free(mem, size);
285 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
286 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
289 vcpu_time = &hv_clock[cpu].pvti;
290 flags = pvclock_read_flags(vcpu_time);
292 kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
295 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
296 x86_platform.calibrate_cpu = kvm_get_tsc_khz;
297 x86_platform.get_wallclock = kvm_get_wallclock;
298 x86_platform.set_wallclock = kvm_set_wallclock;
299 #ifdef CONFIG_X86_LOCAL_APIC
300 x86_cpuinit.early_percpu_clock_init =
301 kvm_setup_secondary_clock;
303 x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
304 x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
305 machine_ops.shutdown = kvm_shutdown;
306 #ifdef CONFIG_KEXEC_CORE
307 machine_ops.crash_shutdown = kvm_crash_shutdown;
309 kvm_get_preset_lpj();
310 clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
311 pv_info.name = "KVM";
314 int __init kvm_setup_vsyscall_timeinfo(void)
319 struct pvclock_vcpu_time_info *vcpu_time;
325 size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
329 vcpu_time = &hv_clock[cpu].pvti;
330 flags = pvclock_read_flags(vcpu_time);
332 if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
339 kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;