xref: /linux/arch/x86/kernel/kvmclock.c (revision d39d0ed196aa1685bb24771e92f78633c66ac9cb)
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 
26 #include <asm/x86_init.h>
27 #include <asm/reboot.h>
28 
29 #define KVM_SCALE 22
30 
31 static int kvmclock = 1;
32 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
33 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
34 
35 static int parse_no_kvmclock(char *arg)
36 {
37 	kvmclock = 0;
38 	return 0;
39 }
40 early_param("no-kvmclock", parse_no_kvmclock);
41 
42 /* The hypervisor will put information about time periodically here */
43 static DEFINE_PER_CPU_SHARED_ALIGNED(struct pvclock_vcpu_time_info, hv_clock);
44 static struct pvclock_wall_clock wall_clock;
45 
46 /*
47  * The wallclock is the time of day when we booted. Since then, some time may
48  * have elapsed since the hypervisor wrote the data. So we try to account for
49  * that with system time
50  */
51 static unsigned long kvm_get_wallclock(void)
52 {
53 	struct pvclock_vcpu_time_info *vcpu_time;
54 	struct timespec ts;
55 	int low, high;
56 
57 	low = (int)__pa_symbol(&wall_clock);
58 	high = ((u64)__pa_symbol(&wall_clock) >> 32);
59 
60 	native_write_msr(msr_kvm_wall_clock, low, high);
61 
62 	vcpu_time = &get_cpu_var(hv_clock);
63 	pvclock_read_wallclock(&wall_clock, vcpu_time, &ts);
64 	put_cpu_var(hv_clock);
65 
66 	return ts.tv_sec;
67 }
68 
69 static int kvm_set_wallclock(unsigned long now)
70 {
71 	return -1;
72 }
73 
74 static cycle_t kvm_clock_read(void)
75 {
76 	struct pvclock_vcpu_time_info *src;
77 	cycle_t ret;
78 
79 	src = &get_cpu_var(hv_clock);
80 	ret = pvclock_clocksource_read(src);
81 	put_cpu_var(hv_clock);
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 static struct clocksource kvm_clock = {
119 	.name = "kvm-clock",
120 	.read = kvm_clock_get_cycles,
121 	.rating = 400,
122 	.mask = CLOCKSOURCE_MASK(64),
123 	.mult = 1 << KVM_SCALE,
124 	.shift = KVM_SCALE,
125 	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
126 };
127 
128 static int kvm_register_clock(char *txt)
129 {
130 	int cpu = smp_processor_id();
131 	int low, high;
132 	low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1;
133 	high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32);
134 	printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
135 	       cpu, high, low, txt);
136 
137 	return native_write_msr_safe(msr_kvm_system_time, low, high);
138 }
139 
140 #ifdef CONFIG_X86_LOCAL_APIC
141 static void __cpuinit kvm_setup_secondary_clock(void)
142 {
143 	/*
144 	 * Now that the first cpu already had this clocksource initialized,
145 	 * we shouldn't fail.
146 	 */
147 	WARN_ON(kvm_register_clock("secondary cpu clock"));
148 	/* ok, done with our trickery, call native */
149 	setup_secondary_APIC_clock();
150 }
151 #endif
152 
153 #ifdef CONFIG_SMP
154 static void __init kvm_smp_prepare_boot_cpu(void)
155 {
156 	WARN_ON(kvm_register_clock("primary cpu clock"));
157 	native_smp_prepare_boot_cpu();
158 }
159 #endif
160 
161 /*
162  * After the clock is registered, the host will keep writing to the
163  * registered memory location. If the guest happens to shutdown, this memory
164  * won't be valid. In cases like kexec, in which you install a new kernel, this
165  * means a random memory location will be kept being written. So before any
166  * kind of shutdown from our side, we unregister the clock by writting anything
167  * that does not have the 'enable' bit set in the msr
168  */
169 #ifdef CONFIG_KEXEC
170 static void kvm_crash_shutdown(struct pt_regs *regs)
171 {
172 	native_write_msr(msr_kvm_system_time, 0, 0);
173 	native_machine_crash_shutdown(regs);
174 }
175 #endif
176 
177 static void kvm_shutdown(void)
178 {
179 	native_write_msr(msr_kvm_system_time, 0, 0);
180 	native_machine_shutdown();
181 }
182 
183 void __init kvmclock_init(void)
184 {
185 	if (!kvm_para_available())
186 		return;
187 
188 	if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
189 		msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
190 		msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
191 	} else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
192 		return;
193 
194 	printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
195 		msr_kvm_system_time, msr_kvm_wall_clock);
196 
197 	if (kvm_register_clock("boot clock"))
198 		return;
199 	pv_time_ops.sched_clock = kvm_clock_read;
200 	x86_platform.calibrate_tsc = kvm_get_tsc_khz;
201 	x86_platform.get_wallclock = kvm_get_wallclock;
202 	x86_platform.set_wallclock = kvm_set_wallclock;
203 #ifdef CONFIG_X86_LOCAL_APIC
204 	x86_cpuinit.setup_percpu_clockev =
205 		kvm_setup_secondary_clock;
206 #endif
207 #ifdef CONFIG_SMP
208 	smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
209 #endif
210 	machine_ops.shutdown  = kvm_shutdown;
211 #ifdef CONFIG_KEXEC
212 	machine_ops.crash_shutdown  = kvm_crash_shutdown;
213 #endif
214 	kvm_get_preset_lpj();
215 	clocksource_register(&kvm_clock);
216 	pv_info.paravirt_enabled = 1;
217 	pv_info.name = "KVM";
218 
219 	if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
220 		pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
221 }
222