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 = 1;
33 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
34 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
36 static int parse_no_kvmclock(char *arg)
41 early_param("no-kvmclock", parse_no_kvmclock);
43 /* The hypervisor will put information about time periodically here */
44 static struct pvclock_vsyscall_time_info *hv_clock;
45 static struct pvclock_wall_clock wall_clock;
48 * The wallclock is the time of day when we booted. Since then, some time may
49 * have elapsed since the hypervisor wrote the data. So we try to account for
50 * that with system time
52 static void kvm_get_wallclock(struct timespec *now)
54 struct pvclock_vcpu_time_info *vcpu_time;
58 low = (int)__pa_symbol(&wall_clock);
59 high = ((u64)__pa_symbol(&wall_clock) >> 32);
61 native_write_msr(msr_kvm_wall_clock, low, high);
65 vcpu_time = &hv_clock[cpu].pvti;
66 pvclock_read_wallclock(&wall_clock, vcpu_time, now);
71 static int kvm_set_wallclock(const struct timespec *now)
76 static cycle_t kvm_clock_read(void)
78 struct pvclock_vcpu_time_info *src;
82 preempt_disable_notrace();
83 cpu = smp_processor_id();
84 src = &hv_clock[cpu].pvti;
85 ret = pvclock_clocksource_read(src);
86 preempt_enable_notrace();
90 static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
92 return kvm_clock_read();
96 * If we don't do that, there is the possibility that the guest
97 * will calibrate under heavy load - thus, getting a lower lpj -
98 * and execute the delays themselves without load. This is wrong,
99 * because no delay loop can finish beforehand.
100 * Any heuristics is subject to fail, because ultimately, a large
101 * poll of guests can be running and trouble each other. So we preset
104 static unsigned long kvm_get_tsc_khz(void)
106 struct pvclock_vcpu_time_info *src;
108 unsigned long tsc_khz;
111 src = &hv_clock[cpu].pvti;
112 tsc_khz = pvclock_tsc_khz(src);
117 static void kvm_get_preset_lpj(void)
122 khz = kvm_get_tsc_khz();
124 lpj = ((u64)khz * 1000);
129 bool kvm_check_and_clear_guest_paused(void)
132 struct pvclock_vcpu_time_info *src;
133 int cpu = smp_processor_id();
138 src = &hv_clock[cpu].pvti;
139 if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
140 src->flags &= ~PVCLOCK_GUEST_STOPPED;
141 pvclock_touch_watchdogs();
148 static struct clocksource kvm_clock = {
150 .read = kvm_clock_get_cycles,
152 .mask = CLOCKSOURCE_MASK(64),
153 .flags = CLOCK_SOURCE_IS_CONTINUOUS,
156 int kvm_register_clock(char *txt)
158 int cpu = smp_processor_id();
160 struct pvclock_vcpu_time_info *src;
165 src = &hv_clock[cpu].pvti;
166 low = (int)slow_virt_to_phys(src) | 1;
167 high = ((u64)slow_virt_to_phys(src) >> 32);
168 ret = native_write_msr_safe(msr_kvm_system_time, low, high);
169 printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
170 cpu, high, low, txt);
175 static void kvm_save_sched_clock_state(void)
179 static void kvm_restore_sched_clock_state(void)
181 kvm_register_clock("primary cpu clock, resume");
184 #ifdef CONFIG_X86_LOCAL_APIC
185 static void kvm_setup_secondary_clock(void)
188 * Now that the first cpu already had this clocksource initialized,
191 WARN_ON(kvm_register_clock("secondary cpu clock"));
196 * After the clock is registered, the host will keep writing to the
197 * registered memory location. If the guest happens to shutdown, this memory
198 * won't be valid. In cases like kexec, in which you install a new kernel, this
199 * means a random memory location will be kept being written. So before any
200 * kind of shutdown from our side, we unregister the clock by writting anything
201 * that does not have the 'enable' bit set in the msr
203 #ifdef CONFIG_KEXEC_CORE
204 static void kvm_crash_shutdown(struct pt_regs *regs)
206 native_write_msr(msr_kvm_system_time, 0, 0);
207 kvm_disable_steal_time();
208 native_machine_crash_shutdown(regs);
212 static void kvm_shutdown(void)
214 native_write_msr(msr_kvm_system_time, 0, 0);
215 kvm_disable_steal_time();
216 native_machine_shutdown();
219 void __init kvmclock_init(void)
221 struct pvclock_vcpu_time_info *vcpu_time;
226 size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
228 if (!kvm_para_available())
231 if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
232 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
233 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
234 } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
237 printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
238 msr_kvm_system_time, msr_kvm_wall_clock);
240 mem = memblock_alloc(size, PAGE_SIZE);
243 hv_clock = __va(mem);
244 memset(hv_clock, 0, size);
246 if (kvm_register_clock("primary cpu clock")) {
248 memblock_free(mem, size);
251 pv_time_ops.sched_clock = kvm_clock_read;
252 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
253 x86_platform.get_wallclock = kvm_get_wallclock;
254 x86_platform.set_wallclock = kvm_set_wallclock;
255 #ifdef CONFIG_X86_LOCAL_APIC
256 x86_cpuinit.early_percpu_clock_init =
257 kvm_setup_secondary_clock;
259 x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
260 x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
261 machine_ops.shutdown = kvm_shutdown;
262 #ifdef CONFIG_KEXEC_CORE
263 machine_ops.crash_shutdown = kvm_crash_shutdown;
265 kvm_get_preset_lpj();
266 clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
267 pv_info.name = "KVM";
269 if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
270 pvclock_set_flags(~0);
273 vcpu_time = &hv_clock[cpu].pvti;
274 flags = pvclock_read_flags(vcpu_time);
275 if (flags & PVCLOCK_COUNTS_FROM_ZERO)
276 set_sched_clock_stable();
280 int __init kvm_setup_vsyscall_timeinfo(void)
286 struct pvclock_vcpu_time_info *vcpu_time;
292 size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
296 vcpu_time = &hv_clock[cpu].pvti;
297 flags = pvclock_read_flags(vcpu_time);
299 if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
304 if ((ret = pvclock_init_vsyscall(hv_clock, size))) {
311 kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;