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>
29 #include <asm/x86_init.h>
30 #include <asm/reboot.h>
32 static int kvmclock __ro_after_init = 1;
33 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
34 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
35 static u64 kvm_sched_clock_offset;
37 static int parse_no_kvmclock(char *arg)
42 early_param("no-kvmclock", parse_no_kvmclock);
44 /* The hypervisor will put information about time periodically here */
45 static struct pvclock_vsyscall_time_info *hv_clock;
46 static struct pvclock_wall_clock wall_clock;
48 struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void)
54 * The wallclock is the time of day when we booted. Since then, some time may
55 * have elapsed since the hypervisor wrote the data. So we try to account for
56 * that with system time
58 static void kvm_get_wallclock(struct timespec *now)
60 struct pvclock_vcpu_time_info *vcpu_time;
64 low = (int)__pa_symbol(&wall_clock);
65 high = ((u64)__pa_symbol(&wall_clock) >> 32);
67 native_write_msr(msr_kvm_wall_clock, low, high);
71 vcpu_time = &hv_clock[cpu].pvti;
72 pvclock_read_wallclock(&wall_clock, vcpu_time, now);
77 static int kvm_set_wallclock(const struct timespec *now)
82 static u64 kvm_clock_read(void)
84 struct pvclock_vcpu_time_info *src;
88 preempt_disable_notrace();
89 cpu = smp_processor_id();
90 src = &hv_clock[cpu].pvti;
91 ret = pvclock_clocksource_read(src);
92 preempt_enable_notrace();
96 static u64 kvm_clock_get_cycles(struct clocksource *cs)
98 return kvm_clock_read();
101 static u64 kvm_sched_clock_read(void)
103 return kvm_clock_read() - kvm_sched_clock_offset;
106 static inline void kvm_sched_clock_init(bool stable)
109 pv_time_ops.sched_clock = kvm_clock_read;
113 kvm_sched_clock_offset = kvm_clock_read();
114 pv_time_ops.sched_clock = kvm_sched_clock_read;
115 set_sched_clock_stable();
117 printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
118 kvm_sched_clock_offset);
120 BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
121 sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
125 * If we don't do that, there is the possibility that the guest
126 * will calibrate under heavy load - thus, getting a lower lpj -
127 * and execute the delays themselves without load. This is wrong,
128 * because no delay loop can finish beforehand.
129 * Any heuristics is subject to fail, because ultimately, a large
130 * poll of guests can be running and trouble each other. So we preset
133 static unsigned long kvm_get_tsc_khz(void)
135 struct pvclock_vcpu_time_info *src;
137 unsigned long tsc_khz;
140 src = &hv_clock[cpu].pvti;
141 tsc_khz = pvclock_tsc_khz(src);
146 static void kvm_get_preset_lpj(void)
151 khz = kvm_get_tsc_khz();
153 lpj = ((u64)khz * 1000);
158 bool kvm_check_and_clear_guest_paused(void)
161 struct pvclock_vcpu_time_info *src;
162 int cpu = smp_processor_id();
167 src = &hv_clock[cpu].pvti;
168 if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
169 src->flags &= ~PVCLOCK_GUEST_STOPPED;
170 pvclock_touch_watchdogs();
177 static struct clocksource kvm_clock = {
179 .read = kvm_clock_get_cycles,
181 .mask = CLOCKSOURCE_MASK(64),
182 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
185 int kvm_register_clock(char *txt)
187 int cpu = smp_processor_id();
189 struct pvclock_vcpu_time_info *src;
194 src = &hv_clock[cpu].pvti;
195 low = (int)slow_virt_to_phys(src) | 1;
196 high = ((u64)slow_virt_to_phys(src) >> 32);
197 ret = native_write_msr_safe(msr_kvm_system_time, low, high);
198 printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
199 cpu, high, low, txt);
204 static void kvm_save_sched_clock_state(void)
208 static void kvm_restore_sched_clock_state(void)
210 kvm_register_clock("primary cpu clock, resume");
213 #ifdef CONFIG_X86_LOCAL_APIC
214 static void kvm_setup_secondary_clock(void)
217 * Now that the first cpu already had this clocksource initialized,
220 WARN_ON(kvm_register_clock("secondary cpu clock"));
225 * After the clock is registered, the host will keep writing to the
226 * registered memory location. If the guest happens to shutdown, this memory
227 * won't be valid. In cases like kexec, in which you install a new kernel, this
228 * means a random memory location will be kept being written. So before any
229 * kind of shutdown from our side, we unregister the clock by writing anything
230 * that does not have the 'enable' bit set in the msr
232 #ifdef CONFIG_KEXEC_CORE
233 static void kvm_crash_shutdown(struct pt_regs *regs)
235 native_write_msr(msr_kvm_system_time, 0, 0);
236 kvm_disable_steal_time();
237 native_machine_crash_shutdown(regs);
241 static void kvm_shutdown(void)
243 native_write_msr(msr_kvm_system_time, 0, 0);
244 kvm_disable_steal_time();
245 native_machine_shutdown();
248 void __init kvmclock_init(void)
250 struct pvclock_vcpu_time_info *vcpu_time;
255 size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
257 if (!kvm_para_available())
260 if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
261 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
262 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
263 } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
266 printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
267 msr_kvm_system_time, msr_kvm_wall_clock);
269 mem = memblock_alloc(size, PAGE_SIZE);
272 hv_clock = __va(mem);
273 memset(hv_clock, 0, size);
275 if (kvm_register_clock("primary cpu clock")) {
277 memblock_free(mem, size);
281 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
282 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
285 vcpu_time = &hv_clock[cpu].pvti;
286 flags = pvclock_read_flags(vcpu_time);
288 kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
291 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
292 x86_platform.calibrate_cpu = kvm_get_tsc_khz;
293 x86_platform.get_wallclock = kvm_get_wallclock;
294 x86_platform.set_wallclock = kvm_set_wallclock;
295 #ifdef CONFIG_X86_LOCAL_APIC
296 x86_cpuinit.early_percpu_clock_init =
297 kvm_setup_secondary_clock;
299 x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
300 x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
301 machine_ops.shutdown = kvm_shutdown;
302 #ifdef CONFIG_KEXEC_CORE
303 machine_ops.crash_shutdown = kvm_crash_shutdown;
305 kvm_get_preset_lpj();
306 clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
307 pv_info.name = "KVM";
310 int __init kvm_setup_vsyscall_timeinfo(void)
315 struct pvclock_vcpu_time_info *vcpu_time;
321 size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
325 vcpu_time = &hv_clock[cpu].pvti;
326 flags = pvclock_read_flags(vcpu_time);
328 if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
335 kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;