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 #define KVM_SCALE 22
  30
  31 static int kvmclock = 1;
  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         native_write_msr(MSR_KVM_WALL_CLOCK, low, high);
  58
  59         vcpu_time = &get_cpu_var(hv_clock);
  60         pvclock_read_wallclock(&wall_clock, vcpu_time, &ts);
  61         put_cpu_var(hv_clock);
  62
  63         return ts.tv_sec;
  64 }
  65
  66 static int kvm_set_wallclock(unsigned long now)
  67 {
  68         return -1;
  69 }
  70
  71 static cycle_t kvm_clock_read(void)
  72 {
  73         struct pvclock_vcpu_time_info *src;
  74         cycle_t ret;
  75
  76         src = &get_cpu_var(hv_clock);
  77         ret = pvclock_clocksource_read(src);
  78         put_cpu_var(hv_clock);
  79         return ret;
  80 }
  81
  82 static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
  83 {
  84         return kvm_clock_read();
  85 }
  86
  87 /*
  88  * If we don't do that, there is the possibility that the guest
  89  * will calibrate under heavy load - thus, getting a lower lpj -
  90  * and execute the delays themselves without load. This is wrong,
  91  * because no delay loop can finish beforehand.
  92  * Any heuristics is subject to fail, because ultimately, a large
  93  * poll of guests can be running and trouble each other. So we preset
  94  * lpj here
  95  */
  96 static unsigned long kvm_get_tsc_khz(void)
  97 {
  98         struct pvclock_vcpu_time_info *src;
  99         src = &per_cpu(hv_clock, 0);
 100         return pvclock_tsc_khz(src);
 101 }
 102
 103 static void kvm_get_preset_lpj(void)
 104 {
 105         unsigned long khz;
 106         u64 lpj;
 107
 108         khz = kvm_get_tsc_khz();
 109
 110         lpj = ((u64)khz * 1000);
 111         do_div(lpj, HZ);
 112         preset_lpj = lpj;
 113 }
 114
 115 static struct clocksource kvm_clock = {
 116         .name = "kvm-clock",
 117         .read = kvm_clock_get_cycles,
 118         .rating = 400,
 119         .mask = CLOCKSOURCE_MASK(64),
 120         .mult = 1 << KVM_SCALE,
 121         .shift = KVM_SCALE,
 122         .flags = CLOCK_SOURCE_IS_CONTINUOUS,
 123 };
 124
 125 static int kvm_register_clock(char *txt)
 126 {
 127         int cpu = smp_processor_id();
 128         int low, high;
 129         low = (int)__pa(&per_cpu(hv_clock, cpu)) | 1;
 130         high = ((u64)__pa(&per_cpu(hv_clock, cpu)) >> 32);
 131         printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
 132                cpu, high, low, txt);
 133         return native_write_msr_safe(MSR_KVM_SYSTEM_TIME, low, high);
 134 }
 135
 136 #ifdef CONFIG_X86_LOCAL_APIC
 137 static void __cpuinit kvm_setup_secondary_clock(void)
 138 {
 139         /*
 140          * Now that the first cpu already had this clocksource initialized,
 141          * we shouldn't fail.
 142          */
 143         WARN_ON(kvm_register_clock("secondary cpu clock"));
 144         /* ok, done with our trickery, call native */
 145         setup_secondary_APIC_clock();
 146 }
 147 #endif
 148
 149 #ifdef CONFIG_SMP
 150 static void __init kvm_smp_prepare_boot_cpu(void)
 151 {
 152         WARN_ON(kvm_register_clock("primary cpu clock"));
 153         native_smp_prepare_boot_cpu();
 154 }
 155 #endif
 156
 157 /*
 158  * After the clock is registered, the host will keep writing to the
 159  * registered memory location. If the guest happens to shutdown, this memory
 160  * won't be valid. In cases like kexec, in which you install a new kernel, this
 161  * means a random memory location will be kept being written. So before any
 162  * kind of shutdown from our side, we unregister the clock by writting anything
 163  * that does not have the 'enable' bit set in the msr
 164  */
 165 #ifdef CONFIG_KEXEC
 166 static void kvm_crash_shutdown(struct pt_regs *regs)
 167 {
 168         native_write_msr_safe(MSR_KVM_SYSTEM_TIME, 0, 0);
 169         native_machine_crash_shutdown(regs);
 170 }
 171 #endif
 172
 173 static void kvm_shutdown(void)
 174 {
 175         native_write_msr_safe(MSR_KVM_SYSTEM_TIME, 0, 0);
 176         native_machine_shutdown();
 177 }
 178
 179 void __init kvmclock_init(void)
 180 {
 181         if (!kvm_para_available())
 182                 return;
 183
 184         if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)) {
 185                 if (kvm_register_clock("boot clock"))
 186                         return;
 187                 pv_time_ops.sched_clock = kvm_clock_read;
 188                 x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 189                 x86_platform.get_wallclock = kvm_get_wallclock;
 190                 x86_platform.set_wallclock = kvm_set_wallclock;
 191 #ifdef CONFIG_X86_LOCAL_APIC
 192                 x86_cpuinit.setup_percpu_clockev =
 193                         kvm_setup_secondary_clock;
 194 #endif
 195 #ifdef CONFIG_SMP
 196                 smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
 197 #endif
 198                 machine_ops.shutdown  = kvm_shutdown;
 199 #ifdef CONFIG_KEXEC
 200                 machine_ops.crash_shutdown  = kvm_crash_shutdown;
 201 #endif
 202                 kvm_get_preset_lpj();
 203                 clocksource_register(&kvm_clock);
 204                 pv_info.paravirt_enabled = 1;
 205                 pv_info.name = "KVM";
 206         }
 207 }