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/sched.h>
27 #include <linux/sched/clock.h>
30 #include <asm/mem_encrypt.h>
31 #include <asm/x86_init.h>
32 #include <asm/reboot.h>
33 #include <asm/kvmclock.h>
35 static int kvmclock __ro_after_init = 1;
36 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
37 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
38 static u64 kvm_sched_clock_offset;
40 static int parse_no_kvmclock(char *arg)
45 early_param("no-kvmclock", parse_no_kvmclock);
47 /* Aligned to page sizes to match whats mapped via vsyscalls to userspace */
48 #define HV_CLOCK_SIZE (sizeof(struct pvclock_vsyscall_time_info) * NR_CPUS)
50 static u8 hv_clock_mem[PAGE_ALIGN(HV_CLOCK_SIZE)] __aligned(PAGE_SIZE);
52 /* The hypervisor will put information about time periodically here */
53 static struct pvclock_vsyscall_time_info *hv_clock;
54 static struct pvclock_wall_clock wall_clock;
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 timespec64 *now)
63 struct pvclock_vcpu_time_info *vcpu_time;
67 low = (int)slow_virt_to_phys(&wall_clock);
68 high = ((u64)slow_virt_to_phys(&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 timespec64 *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);
146 setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
150 static void kvm_get_preset_lpj(void)
155 khz = kvm_get_tsc_khz();
157 lpj = ((u64)khz * 1000);
162 bool kvm_check_and_clear_guest_paused(void)
165 struct pvclock_vcpu_time_info *src;
166 int cpu = smp_processor_id();
171 src = &hv_clock[cpu].pvti;
172 if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
173 src->flags &= ~PVCLOCK_GUEST_STOPPED;
174 pvclock_touch_watchdogs();
181 struct clocksource kvm_clock = {
183 .read = kvm_clock_get_cycles,
185 .mask = CLOCKSOURCE_MASK(64),
186 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
188 EXPORT_SYMBOL_GPL(kvm_clock);
190 static void kvm_register_clock(char *txt)
192 struct pvclock_vcpu_time_info *src;
193 int cpu = smp_processor_id();
199 src = &hv_clock[cpu].pvti;
200 pa = slow_virt_to_phys(src) | 0x01ULL;
201 wrmsrl(msr_kvm_system_time, pa);
202 pr_info("kvm-clock: cpu %d, msr %llx, %s\n", cpu, pa, txt);
205 static void kvm_save_sched_clock_state(void)
209 static void kvm_restore_sched_clock_state(void)
211 kvm_register_clock("primary cpu clock, resume");
214 #ifdef CONFIG_X86_LOCAL_APIC
215 static void kvm_setup_secondary_clock(void)
217 kvm_register_clock("secondary cpu clock");
222 * After the clock is registered, the host will keep writing to the
223 * registered memory location. If the guest happens to shutdown, this memory
224 * won't be valid. In cases like kexec, in which you install a new kernel, this
225 * means a random memory location will be kept being written. So before any
226 * kind of shutdown from our side, we unregister the clock by writing anything
227 * that does not have the 'enable' bit set in the msr
229 #ifdef CONFIG_KEXEC_CORE
230 static void kvm_crash_shutdown(struct pt_regs *regs)
232 native_write_msr(msr_kvm_system_time, 0, 0);
233 kvm_disable_steal_time();
234 native_machine_crash_shutdown(regs);
238 static void kvm_shutdown(void)
240 native_write_msr(msr_kvm_system_time, 0, 0);
241 kvm_disable_steal_time();
242 native_machine_shutdown();
245 void __init kvmclock_init(void)
247 struct pvclock_vcpu_time_info *vcpu_time;
251 if (!kvm_para_available())
254 if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
255 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
256 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
257 } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
260 printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
261 msr_kvm_system_time, msr_kvm_wall_clock);
263 hv_clock = (struct pvclock_vsyscall_time_info *)hv_clock_mem;
264 kvm_register_clock("primary cpu clock");
265 pvclock_set_pvti_cpu0_va(hv_clock);
267 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
268 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
271 vcpu_time = &hv_clock[cpu].pvti;
272 flags = pvclock_read_flags(vcpu_time);
274 kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
277 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
278 x86_platform.calibrate_cpu = kvm_get_tsc_khz;
279 x86_platform.get_wallclock = kvm_get_wallclock;
280 x86_platform.set_wallclock = kvm_set_wallclock;
281 #ifdef CONFIG_X86_LOCAL_APIC
282 x86_cpuinit.early_percpu_clock_init =
283 kvm_setup_secondary_clock;
285 x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
286 x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
287 machine_ops.shutdown = kvm_shutdown;
288 #ifdef CONFIG_KEXEC_CORE
289 machine_ops.crash_shutdown = kvm_crash_shutdown;
291 kvm_get_preset_lpj();
292 clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
293 pv_info.name = "KVM";
296 int __init kvm_setup_vsyscall_timeinfo(void)
301 struct pvclock_vcpu_time_info *vcpu_time;
308 vcpu_time = &hv_clock[cpu].pvti;
309 flags = pvclock_read_flags(vcpu_time);
313 if (!(flags & PVCLOCK_TSC_STABLE_BIT))
316 kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;