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         preempt_disable_notrace();
  78         src = &__get_cpu_var(hv_clock);
  79         ret = pvclock_clocksource_read(src);
  80         preempt_enable_notrace();
  81         return ret;
  82 }
  83
  84 static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
  85 {
  86         return kvm_clock_read();
  87 }
  88
  89 /*
  90  * If we don't do that, there is the possibility that the guest
  91  * will calibrate under heavy load - thus, getting a lower lpj -
  92  * and execute the delays themselves without load. This is wrong,
  93  * because no delay loop can finish beforehand.
  94  * Any heuristics is subject to fail, because ultimately, a large
  95  * poll of guests can be running and trouble each other. So we preset
  96  * lpj here
  97  */
  98 static unsigned long kvm_get_tsc_khz(void)
  99 {
 100         struct pvclock_vcpu_time_info *src;
 101         src = &per_cpu(hv_clock, 0);
 102         return pvclock_tsc_khz(src);
 103 }
 104
 105 static void kvm_get_preset_lpj(void)
 106 {
 107         unsigned long khz;
 108         u64 lpj;
 109
 110         khz = kvm_get_tsc_khz();
 111
 112         lpj = ((u64)khz * 1000);
 113         do_div(lpj, HZ);
 114         preset_lpj = lpj;
 115 }
 116
 117 static struct clocksource kvm_clock = {
 118         .name = "kvm-clock",
 119         .read = kvm_clock_get_cycles,
 120         .rating = 400,
 121         .mask = CLOCKSOURCE_MASK(64),
 122         .flags = CLOCK_SOURCE_IS_CONTINUOUS,
 123 };
 124
 125 int kvm_register_clock(char *txt)
 126 {
 127         int cpu = smp_processor_id();
 128         int low, high, ret;
 129
 130         low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1;
 131         high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32);
 132         ret = native_write_msr_safe(msr_kvm_system_time, low, high);
 133         printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
 134                cpu, high, low, txt);
 135
 136         return ret;
 137 }
 138
 139 #ifdef CONFIG_X86_LOCAL_APIC
 140 static void __cpuinit kvm_setup_secondary_clock(void)
 141 {
 142         /*
 143          * Now that the first cpu already had this clocksource initialized,
 144          * we shouldn't fail.
 145          */
 146         WARN_ON(kvm_register_clock("secondary cpu clock"));
 147         /* ok, done with our trickery, call native */
 148         setup_secondary_APIC_clock();
 149 }
 150 #endif
 151
 152 /*
 153  * After the clock is registered, the host will keep writing to the
 154  * registered memory location. If the guest happens to shutdown, this memory
 155  * won't be valid. In cases like kexec, in which you install a new kernel, this
 156  * means a random memory location will be kept being written. So before any
 157  * kind of shutdown from our side, we unregister the clock by writting anything
 158  * that does not have the 'enable' bit set in the msr
 159  */
 160 #ifdef CONFIG_KEXEC
 161 static void kvm_crash_shutdown(struct pt_regs *regs)
 162 {
 163         native_write_msr(msr_kvm_system_time, 0, 0);
 164         kvm_disable_steal_time();
 165         native_machine_crash_shutdown(regs);
 166 }
 167 #endif
 168
 169 static void kvm_shutdown(void)
 170 {
 171         native_write_msr(msr_kvm_system_time, 0, 0);
 172         kvm_disable_steal_time();
 173         native_machine_shutdown();
 174 }
 175
 176 void __init kvmclock_init(void)
 177 {
 178         if (!kvm_para_available())
 179                 return;
 180
 181         if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
 182                 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
 183                 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
 184         } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
 185                 return;
 186
 187         printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
 188                 msr_kvm_system_time, msr_kvm_wall_clock);
 189
 190         if (kvm_register_clock("boot clock"))
 191                 return;
 192         pv_time_ops.sched_clock = kvm_clock_read;
 193         x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 194         x86_platform.get_wallclock = kvm_get_wallclock;
 195         x86_platform.set_wallclock = kvm_set_wallclock;
 196 #ifdef CONFIG_X86_LOCAL_APIC
 197         x86_cpuinit.setup_percpu_clockev =
 198                 kvm_setup_secondary_clock;
 199 #endif
 200         machine_ops.shutdown  = kvm_shutdown;
 201 #ifdef CONFIG_KEXEC
 202         machine_ops.crash_shutdown  = kvm_crash_shutdown;
 203 #endif
 204         kvm_get_preset_lpj();
 205         clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
 206         pv_info.paravirt_enabled = 1;
 207         pv_info.name = "KVM";
 208
 209         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 210                 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 211 }