arch/x86/kernel/kvmclock.c

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