xref: /linux/arch/x86/kernel/kvmclock.c (revision ee665ecca6d6775f65b1a4154c34f551f62cec52)
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 <asm/reboot.h>
26 
27 #define KVM_SCALE 22
28 
29 static int kvmclock = 1;
30 
31 static int parse_no_kvmclock(char *arg)
32 {
33 	kvmclock = 0;
34 	return 0;
35 }
36 early_param("no-kvmclock", parse_no_kvmclock);
37 
38 /* The hypervisor will put information about time periodically here */
39 static DEFINE_PER_CPU_SHARED_ALIGNED(struct pvclock_vcpu_time_info, hv_clock);
40 static struct pvclock_wall_clock wall_clock;
41 
42 /*
43  * The wallclock is the time of day when we booted. Since then, some time may
44  * have elapsed since the hypervisor wrote the data. So we try to account for
45  * that with system time
46  */
47 static unsigned long kvm_get_wallclock(void)
48 {
49 	struct pvclock_vcpu_time_info *vcpu_time;
50 	struct timespec ts;
51 	int low, high;
52 
53 	low = (int)__pa(&wall_clock);
54 	high = ((u64)__pa(&wall_clock) >> 32);
55 	native_write_msr(MSR_KVM_WALL_CLOCK, low, high);
56 
57 	vcpu_time = &get_cpu_var(hv_clock);
58 	pvclock_read_wallclock(&wall_clock, vcpu_time, &ts);
59 	put_cpu_var(hv_clock);
60 
61 	return ts.tv_sec;
62 }
63 
64 static int kvm_set_wallclock(unsigned long now)
65 {
66 	return -1;
67 }
68 
69 static cycle_t kvm_clock_read(void)
70 {
71 	struct pvclock_vcpu_time_info *src;
72 	cycle_t ret;
73 
74 	src = &get_cpu_var(hv_clock);
75 	ret = pvclock_clocksource_read(src);
76 	put_cpu_var(hv_clock);
77 	return ret;
78 }
79 
80 /*
81  * If we don't do that, there is the possibility that the guest
82  * will calibrate under heavy load - thus, getting a lower lpj -
83  * and execute the delays themselves without load. This is wrong,
84  * because no delay loop can finish beforehand.
85  * Any heuristics is subject to fail, because ultimately, a large
86  * poll of guests can be running and trouble each other. So we preset
87  * lpj here
88  */
89 static unsigned long kvm_get_tsc_khz(void)
90 {
91 	struct pvclock_vcpu_time_info *src;
92 	src = &per_cpu(hv_clock, 0);
93 	return pvclock_tsc_khz(src);
94 }
95 
96 static void kvm_get_preset_lpj(void)
97 {
98 	unsigned long khz;
99 	u64 lpj;
100 
101 	khz = kvm_get_tsc_khz();
102 
103 	lpj = ((u64)khz * 1000);
104 	do_div(lpj, HZ);
105 	preset_lpj = lpj;
106 }
107 
108 static struct clocksource kvm_clock = {
109 	.name = "kvm-clock",
110 	.read = kvm_clock_read,
111 	.rating = 400,
112 	.mask = CLOCKSOURCE_MASK(64),
113 	.mult = 1 << KVM_SCALE,
114 	.shift = KVM_SCALE,
115 	.flags = CLOCK_SOURCE_IS_CONTINUOUS,
116 };
117 
118 static int kvm_register_clock(char *txt)
119 {
120 	int cpu = smp_processor_id();
121 	int low, high;
122 	low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1;
123 	high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32);
124 	printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
125 	       cpu, high, low, txt);
126 	return native_write_msr_safe(MSR_KVM_SYSTEM_TIME, low, high);
127 }
128 
129 #ifdef CONFIG_X86_LOCAL_APIC
130 static void __cpuinit kvm_setup_secondary_clock(void)
131 {
132 	/*
133 	 * Now that the first cpu already had this clocksource initialized,
134 	 * we shouldn't fail.
135 	 */
136 	WARN_ON(kvm_register_clock("secondary cpu clock"));
137 	/* ok, done with our trickery, call native */
138 	setup_secondary_APIC_clock();
139 }
140 #endif
141 
142 #ifdef CONFIG_SMP
143 static void __init kvm_smp_prepare_boot_cpu(void)
144 {
145 	WARN_ON(kvm_register_clock("primary cpu clock"));
146 	native_smp_prepare_boot_cpu();
147 }
148 #endif
149 
150 /*
151  * After the clock is registered, the host will keep writing to the
152  * registered memory location. If the guest happens to shutdown, this memory
153  * won't be valid. In cases like kexec, in which you install a new kernel, this
154  * means a random memory location will be kept being written. So before any
155  * kind of shutdown from our side, we unregister the clock by writting anything
156  * that does not have the 'enable' bit set in the msr
157  */
158 #ifdef CONFIG_KEXEC
159 static void kvm_crash_shutdown(struct pt_regs *regs)
160 {
161 	native_write_msr_safe(MSR_KVM_SYSTEM_TIME, 0, 0);
162 	native_machine_crash_shutdown(regs);
163 }
164 #endif
165 
166 static void kvm_shutdown(void)
167 {
168 	native_write_msr_safe(MSR_KVM_SYSTEM_TIME, 0, 0);
169 	native_machine_shutdown();
170 }
171 
172 void __init kvmclock_init(void)
173 {
174 	if (!kvm_para_available())
175 		return;
176 
177 	if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)) {
178 		if (kvm_register_clock("boot clock"))
179 			return;
180 		pv_time_ops.get_wallclock = kvm_get_wallclock;
181 		pv_time_ops.set_wallclock = kvm_set_wallclock;
182 		pv_time_ops.sched_clock = kvm_clock_read;
183 		pv_time_ops.get_tsc_khz = kvm_get_tsc_khz;
184 #ifdef CONFIG_X86_LOCAL_APIC
185 		pv_apic_ops.setup_secondary_clock = kvm_setup_secondary_clock;
186 #endif
187 #ifdef CONFIG_SMP
188 		smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
189 #endif
190 		machine_ops.shutdown  = kvm_shutdown;
191 #ifdef CONFIG_KEXEC
192 		machine_ops.crash_shutdown  = kvm_crash_shutdown;
193 #endif
194 		kvm_get_preset_lpj();
195 		clocksource_register(&kvm_clock);
196 		pv_info.paravirt_enabled = 1;
197 		pv_info.name = "KVM";
198 	}
199 }
200