xref: /linux/arch/x86/kernel/kvmclock.c (revision 4413e16d9d21673bb5048a2e542f1aaa00015c2e)
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>
21 #include <asm/pvclock.h>
22 #include <asm/msr.h>
23 #include <asm/apic.h>
24 #include <linux/percpu.h>
25 #include <linux/hardirq.h>
26 
27 #include <asm/x86_init.h>
28 #include <asm/reboot.h>
29 
30 static int kvmclock = 1;
31 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
32 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
33 
34 static int parse_no_kvmclock(char *arg)
35 {
36 	kvmclock = 0;
37 	return 0;
38 }
39 early_param("no-kvmclock", parse_no_kvmclock);
40 
41 /* The hypervisor will put information about time periodically here */
42 static DEFINE_PER_CPU_SHARED_ALIGNED(struct pvclock_vcpu_time_info, hv_clock);
43 static struct pvclock_wall_clock wall_clock;
44 
45 /*
46  * The wallclock is the time of day when we booted. Since then, some time may
47  * have elapsed since the hypervisor wrote the data. So we try to account for
48  * that with system time
49  */
50 static unsigned long kvm_get_wallclock(void)
51 {
52 	struct pvclock_vcpu_time_info *vcpu_time;
53 	struct timespec ts;
54 	int low, high;
55 
56 	low = (int)__pa_symbol(&wall_clock);
57 	high = ((u64)__pa_symbol(&wall_clock) >> 32);
58 
59 	native_write_msr(msr_kvm_wall_clock, low, high);
60 
61 	vcpu_time = &get_cpu_var(hv_clock);
62 	pvclock_read_wallclock(&wall_clock, vcpu_time, &ts);
63 	put_cpu_var(hv_clock);
64 
65 	return ts.tv_sec;
66 }
67 
68 static int kvm_set_wallclock(unsigned long now)
69 {
70 	return -1;
71 }
72 
73 static cycle_t kvm_clock_read(void)
74 {
75 	struct pvclock_vcpu_time_info *src;
76 	cycle_t ret;
77 
78 	preempt_disable_notrace();
79 	src = &__get_cpu_var(hv_clock);
80 	ret = pvclock_clocksource_read(src);
81 	preempt_enable_notrace();
82 	return ret;
83 }
84 
85 static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
86 {
87 	return kvm_clock_read();
88 }
89 
90 /*
91  * If we don't do that, there is the possibility that the guest
92  * will calibrate under heavy load - thus, getting a lower lpj -
93  * and execute the delays themselves without load. This is wrong,
94  * because no delay loop can finish beforehand.
95  * Any heuristics is subject to fail, because ultimately, a large
96  * poll of guests can be running and trouble each other. So we preset
97  * lpj here
98  */
99 static unsigned long kvm_get_tsc_khz(void)
100 {
101 	struct pvclock_vcpu_time_info *src;
102 	src = &per_cpu(hv_clock, 0);
103 	return pvclock_tsc_khz(src);
104 }
105 
106 static void kvm_get_preset_lpj(void)
107 {
108 	unsigned long khz;
109 	u64 lpj;
110 
111 	khz = kvm_get_tsc_khz();
112 
113 	lpj = ((u64)khz * 1000);
114 	do_div(lpj, HZ);
115 	preset_lpj = lpj;
116 }
117 
118 bool kvm_check_and_clear_guest_paused(void)
119 {
120 	bool ret = false;
121 	struct pvclock_vcpu_time_info *src;
122 
123 	src = &__get_cpu_var(hv_clock);
124 	if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
125 		__this_cpu_and(hv_clock.flags, ~PVCLOCK_GUEST_STOPPED);
126 		ret = true;
127 	}
128 
129 	return ret;
130 }
131 
132 static struct clocksource kvm_clock = {
133 	.name = "kvm-clock",
134 	.read = kvm_clock_get_cycles,
135 	.rating = 400,
136 	.mask = CLOCKSOURCE_MASK(64),
137 	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
138 };
139 
140 int kvm_register_clock(char *txt)
141 {
142 	int cpu = smp_processor_id();
143 	int low, high, ret;
144 
145 	low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1;
146 	high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32);
147 	ret = native_write_msr_safe(msr_kvm_system_time, low, high);
148 	printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
149 	       cpu, high, low, txt);
150 
151 	return ret;
152 }
153 
154 static void kvm_save_sched_clock_state(void)
155 {
156 }
157 
158 static void kvm_restore_sched_clock_state(void)
159 {
160 	kvm_register_clock("primary cpu clock, resume");
161 }
162 
163 #ifdef CONFIG_X86_LOCAL_APIC
164 static void __cpuinit kvm_setup_secondary_clock(void)
165 {
166 	/*
167 	 * Now that the first cpu already had this clocksource initialized,
168 	 * we shouldn't fail.
169 	 */
170 	WARN_ON(kvm_register_clock("secondary cpu clock"));
171 }
172 #endif
173 
174 /*
175  * After the clock is registered, the host will keep writing to the
176  * registered memory location. If the guest happens to shutdown, this memory
177  * won't be valid. In cases like kexec, in which you install a new kernel, this
178  * means a random memory location will be kept being written. So before any
179  * kind of shutdown from our side, we unregister the clock by writting anything
180  * that does not have the 'enable' bit set in the msr
181  */
182 #ifdef CONFIG_KEXEC
183 static void kvm_crash_shutdown(struct pt_regs *regs)
184 {
185 	native_write_msr(msr_kvm_system_time, 0, 0);
186 	kvm_disable_steal_time();
187 	native_machine_crash_shutdown(regs);
188 }
189 #endif
190 
191 static void kvm_shutdown(void)
192 {
193 	native_write_msr(msr_kvm_system_time, 0, 0);
194 	kvm_disable_steal_time();
195 	native_machine_shutdown();
196 }
197 
198 void __init kvmclock_init(void)
199 {
200 	if (!kvm_para_available())
201 		return;
202 
203 	if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
204 		msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
205 		msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
206 	} else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
207 		return;
208 
209 	printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
210 		msr_kvm_system_time, msr_kvm_wall_clock);
211 
212 	if (kvm_register_clock("boot clock"))
213 		return;
214 	pv_time_ops.sched_clock = kvm_clock_read;
215 	x86_platform.calibrate_tsc = kvm_get_tsc_khz;
216 	x86_platform.get_wallclock = kvm_get_wallclock;
217 	x86_platform.set_wallclock = kvm_set_wallclock;
218 #ifdef CONFIG_X86_LOCAL_APIC
219 	x86_cpuinit.early_percpu_clock_init =
220 		kvm_setup_secondary_clock;
221 #endif
222 	x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
223 	x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
224 	machine_ops.shutdown  = kvm_shutdown;
225 #ifdef CONFIG_KEXEC
226 	machine_ops.crash_shutdown  = kvm_crash_shutdown;
227 #endif
228 	kvm_get_preset_lpj();
229 	clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
230 	pv_info.paravirt_enabled = 1;
231 	pv_info.name = "KVM";
232 
233 	if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
234 		pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
235 }
236