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 #include <linux/hardirq.h>
  26 #include <linux/memblock.h>
  27
  28 #include <asm/x86_init.h>
  29 #include <asm/reboot.h>
  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 struct pvclock_vsyscall_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         int cpu;
  57
  58         low = (int)__pa_symbol(&wall_clock);
  59         high = ((u64)__pa_symbol(&wall_clock) >> 32);
  60
  61         native_write_msr(msr_kvm_wall_clock, low, high);
  62
  63         preempt_disable();
  64         cpu = smp_processor_id();
  65
  66         vcpu_time = &hv_clock[cpu].pvti;
  67         pvclock_read_wallclock(&wall_clock, vcpu_time, &ts);
  68
  69         preempt_enable();
  70
  71         return ts.tv_sec;
  72 }
  73
  74 static int kvm_set_wallclock(unsigned long now)
  75 {
  76         return -1;
  77 }
  78
  79 static cycle_t kvm_clock_read(void)
  80 {
  81         struct pvclock_vcpu_time_info *src;
  82         cycle_t ret;
  83         int cpu;
  84
  85         preempt_disable_notrace();
  86         cpu = smp_processor_id();
  87         src = &hv_clock[cpu].pvti;
  88         ret = pvclock_clocksource_read(src);
  89         preempt_enable_notrace();
  90         return ret;
  91 }
  92
  93 static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
  94 {
  95         return kvm_clock_read();
  96 }
  97
  98 /*
  99  * If we don't do that, there is the possibility that the guest
 100  * will calibrate under heavy load - thus, getting a lower lpj -
 101  * and execute the delays themselves without load. This is wrong,
 102  * because no delay loop can finish beforehand.
 103  * Any heuristics is subject to fail, because ultimately, a large
 104  * poll of guests can be running and trouble each other. So we preset
 105  * lpj here
 106  */
 107 static unsigned long kvm_get_tsc_khz(void)
 108 {
 109         struct pvclock_vcpu_time_info *src;
 110         int cpu;
 111         unsigned long tsc_khz;
 112
 113         preempt_disable();
 114         cpu = smp_processor_id();
 115         src = &hv_clock[cpu].pvti;
 116         tsc_khz = pvclock_tsc_khz(src);
 117         preempt_enable();
 118         return tsc_khz;
 119 }
 120
 121 static void kvm_get_preset_lpj(void)
 122 {
 123         unsigned long khz;
 124         u64 lpj;
 125
 126         khz = kvm_get_tsc_khz();
 127
 128         lpj = ((u64)khz * 1000);
 129         do_div(lpj, HZ);
 130         preset_lpj = lpj;
 131 }
 132
 133 bool kvm_check_and_clear_guest_paused(void)
 134 {
 135         bool ret = false;
 136         struct pvclock_vcpu_time_info *src;
 137         int cpu = smp_processor_id();
 138
 139         if (!hv_clock)
 140                 return ret;
 141
 142         src = &hv_clock[cpu].pvti;
 143         if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
 144                 src->flags &= ~PVCLOCK_GUEST_STOPPED;
 145                 ret = true;
 146         }
 147
 148         return ret;
 149 }
 150
 151 static struct clocksource kvm_clock = {
 152         .name = "kvm-clock",
 153         .read = kvm_clock_get_cycles,
 154         .rating = 400,
 155         .mask = CLOCKSOURCE_MASK(64),
 156         .flags = CLOCK_SOURCE_IS_CONTINUOUS,
 157 };
 158
 159 int kvm_register_clock(char *txt)
 160 {
 161         int cpu = smp_processor_id();
 162         int low, high, ret;
 163         struct pvclock_vcpu_time_info *src;
 164
 165         if (!hv_clock)
 166                 return 0;
 167
 168         src = &hv_clock[cpu].pvti;
 169         low = (int)__pa(src) | 1;
 170         high = ((u64)__pa(src) >> 32);
 171         ret = native_write_msr_safe(msr_kvm_system_time, low, high);
 172         printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
 173                cpu, high, low, txt);
 174
 175         return ret;
 176 }
 177
 178 static void kvm_save_sched_clock_state(void)
 179 {
 180 }
 181
 182 static void kvm_restore_sched_clock_state(void)
 183 {
 184         kvm_register_clock("primary cpu clock, resume");
 185 }
 186
 187 #ifdef CONFIG_X86_LOCAL_APIC
 188 static void __cpuinit kvm_setup_secondary_clock(void)
 189 {
 190         /*
 191          * Now that the first cpu already had this clocksource initialized,
 192          * we shouldn't fail.
 193          */
 194         WARN_ON(kvm_register_clock("secondary cpu clock"));
 195 }
 196 #endif
 197
 198 /*
 199  * After the clock is registered, the host will keep writing to the
 200  * registered memory location. If the guest happens to shutdown, this memory
 201  * won't be valid. In cases like kexec, in which you install a new kernel, this
 202  * means a random memory location will be kept being written. So before any
 203  * kind of shutdown from our side, we unregister the clock by writting anything
 204  * that does not have the 'enable' bit set in the msr
 205  */
 206 #ifdef CONFIG_KEXEC
 207 static void kvm_crash_shutdown(struct pt_regs *regs)
 208 {
 209         native_write_msr(msr_kvm_system_time, 0, 0);
 210         kvm_disable_steal_time();
 211         native_machine_crash_shutdown(regs);
 212 }
 213 #endif
 214
 215 static void kvm_shutdown(void)
 216 {
 217         native_write_msr(msr_kvm_system_time, 0, 0);
 218         kvm_disable_steal_time();
 219         native_machine_shutdown();
 220 }
 221
 222 void __init kvmclock_init(void)
 223 {
 224         unsigned long mem;
 225         int size;
 226
 227         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 228
 229         if (!kvm_para_available())
 230                 return;
 231
 232         if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
 233                 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
 234                 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
 235         } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
 236                 return;
 237
 238         printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
 239                 msr_kvm_system_time, msr_kvm_wall_clock);
 240
 241         mem = memblock_alloc(size, PAGE_SIZE);
 242         if (!mem)
 243                 return;
 244         hv_clock = __va(mem);
 245
 246         if (kvm_register_clock("boot clock")) {
 247                 hv_clock = NULL;
 248                 memblock_free(mem, size);
 249                 return;
 250         }
 251         pv_time_ops.sched_clock = kvm_clock_read;
 252         x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 253         x86_platform.get_wallclock = kvm_get_wallclock;
 254         x86_platform.set_wallclock = kvm_set_wallclock;
 255 #ifdef CONFIG_X86_LOCAL_APIC
 256         x86_cpuinit.early_percpu_clock_init =
 257                 kvm_setup_secondary_clock;
 258 #endif
 259         x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
 260         x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
 261         machine_ops.shutdown  = kvm_shutdown;
 262 #ifdef CONFIG_KEXEC
 263         machine_ops.crash_shutdown  = kvm_crash_shutdown;
 264 #endif
 265         kvm_get_preset_lpj();
 266         clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
 267         pv_info.paravirt_enabled = 1;
 268         pv_info.name = "KVM";
 269
 270         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 271                 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 272 }
 273
 274 int __init kvm_setup_vsyscall_timeinfo(void)
 275 {
 276 #ifdef CONFIG_X86_64
 277         int cpu;
 278         int ret;
 279         u8 flags;
 280         struct pvclock_vcpu_time_info *vcpu_time;
 281         unsigned int size;
 282
 283         if (!hv_clock)
 284                 return 0;
 285
 286         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 287
 288         preempt_disable();
 289         cpu = smp_processor_id();
 290
 291         vcpu_time = &hv_clock[cpu].pvti;
 292         flags = pvclock_read_flags(vcpu_time);
 293
 294         if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
 295                 preempt_enable();
 296                 return 1;
 297         }
 298
 299         if ((ret = pvclock_init_vsyscall(hv_clock, size))) {
 300                 preempt_enable();
 301                 return ret;
 302         }
 303
 304         preempt_enable();
 305
 306         kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
 307 #endif
 308         return 0;
 309 }