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 #include <linux/sched.h>
  28
  29 #include <asm/x86_init.h>
  30 #include <asm/reboot.h>
  31
  32 static int kvmclock = 1;
  33 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
  34 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
  35 static cycle_t kvm_sched_clock_offset;
  36
  37 static int parse_no_kvmclock(char *arg)
  38 {
  39         kvmclock = 0;
  40         return 0;
  41 }
  42 early_param("no-kvmclock", parse_no_kvmclock);
  43
  44 /* The hypervisor will put information about time periodically here */
  45 static struct pvclock_vsyscall_time_info *hv_clock;
  46 static struct pvclock_wall_clock wall_clock;
  47
  48 struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void)
  49 {
  50         return hv_clock;
  51 }
  52
  53 /*
  54  * The wallclock is the time of day when we booted. Since then, some time may
  55  * have elapsed since the hypervisor wrote the data. So we try to account for
  56  * that with system time
  57  */
  58 static void kvm_get_wallclock(struct timespec *now)
  59 {
  60         struct pvclock_vcpu_time_info *vcpu_time;
  61         int low, high;
  62         int cpu;
  63
  64         low = (int)__pa_symbol(&wall_clock);
  65         high = ((u64)__pa_symbol(&wall_clock) >> 32);
  66
  67         native_write_msr(msr_kvm_wall_clock, low, high);
  68
  69         cpu = get_cpu();
  70
  71         vcpu_time = &hv_clock[cpu].pvti;
  72         pvclock_read_wallclock(&wall_clock, vcpu_time, now);
  73
  74         put_cpu();
  75 }
  76
  77 static int kvm_set_wallclock(const struct timespec *now)
  78 {
  79         return -1;
  80 }
  81
  82 static cycle_t kvm_clock_read(void)
  83 {
  84         struct pvclock_vcpu_time_info *src;
  85         cycle_t ret;
  86         int cpu;
  87
  88         preempt_disable_notrace();
  89         cpu = smp_processor_id();
  90         src = &hv_clock[cpu].pvti;
  91         ret = pvclock_clocksource_read(src);
  92         preempt_enable_notrace();
  93         return ret;
  94 }
  95
  96 static cycle_t kvm_clock_get_cycles(struct clocksource *cs)
  97 {
  98         return kvm_clock_read();
  99 }
 100
 101 static cycle_t kvm_sched_clock_read(void)
 102 {
 103         return kvm_clock_read() - kvm_sched_clock_offset;
 104 }
 105
 106 static inline void kvm_sched_clock_init(bool stable)
 107 {
 108         if (!stable) {
 109                 pv_time_ops.sched_clock = kvm_clock_read;
 110                 return;
 111         }
 112
 113         kvm_sched_clock_offset = kvm_clock_read();
 114         pv_time_ops.sched_clock = kvm_sched_clock_read;
 115         set_sched_clock_stable();
 116
 117         printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
 118                         kvm_sched_clock_offset);
 119
 120         BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
 121                  sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
 122 }
 123
 124 /*
 125  * If we don't do that, there is the possibility that the guest
 126  * will calibrate under heavy load - thus, getting a lower lpj -
 127  * and execute the delays themselves without load. This is wrong,
 128  * because no delay loop can finish beforehand.
 129  * Any heuristics is subject to fail, because ultimately, a large
 130  * poll of guests can be running and trouble each other. So we preset
 131  * lpj here
 132  */
 133 static unsigned long kvm_get_tsc_khz(void)
 134 {
 135         struct pvclock_vcpu_time_info *src;
 136         int cpu;
 137         unsigned long tsc_khz;
 138
 139         cpu = get_cpu();
 140         src = &hv_clock[cpu].pvti;
 141         tsc_khz = pvclock_tsc_khz(src);
 142         put_cpu();
 143         return tsc_khz;
 144 }
 145
 146 static void kvm_get_preset_lpj(void)
 147 {
 148         unsigned long khz;
 149         u64 lpj;
 150
 151         khz = kvm_get_tsc_khz();
 152
 153         lpj = ((u64)khz * 1000);
 154         do_div(lpj, HZ);
 155         preset_lpj = lpj;
 156 }
 157
 158 bool kvm_check_and_clear_guest_paused(void)
 159 {
 160         bool ret = false;
 161         struct pvclock_vcpu_time_info *src;
 162         int cpu = smp_processor_id();
 163
 164         if (!hv_clock)
 165                 return ret;
 166
 167         src = &hv_clock[cpu].pvti;
 168         if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
 169                 src->flags &= ~PVCLOCK_GUEST_STOPPED;
 170                 pvclock_touch_watchdogs();
 171                 ret = true;
 172         }
 173
 174         return ret;
 175 }
 176
 177 static struct clocksource kvm_clock = {
 178         .name = "kvm-clock",
 179         .read = kvm_clock_get_cycles,
 180         .rating = 400,
 181         .mask = CLOCKSOURCE_MASK(64),
 182         .flags = CLOCK_SOURCE_IS_CONTINUOUS,
 183 };
 184
 185 int kvm_register_clock(char *txt)
 186 {
 187         int cpu = smp_processor_id();
 188         int low, high, ret;
 189         struct pvclock_vcpu_time_info *src;
 190
 191         if (!hv_clock)
 192                 return 0;
 193
 194         src = &hv_clock[cpu].pvti;
 195         low = (int)slow_virt_to_phys(src) | 1;
 196         high = ((u64)slow_virt_to_phys(src) >> 32);
 197         ret = native_write_msr_safe(msr_kvm_system_time, low, high);
 198         printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
 199                cpu, high, low, txt);
 200
 201         return ret;
 202 }
 203
 204 static void kvm_save_sched_clock_state(void)
 205 {
 206 }
 207
 208 static void kvm_restore_sched_clock_state(void)
 209 {
 210         kvm_register_clock("primary cpu clock, resume");
 211 }
 212
 213 #ifdef CONFIG_X86_LOCAL_APIC
 214 static void kvm_setup_secondary_clock(void)
 215 {
 216         /*
 217          * Now that the first cpu already had this clocksource initialized,
 218          * we shouldn't fail.
 219          */
 220         WARN_ON(kvm_register_clock("secondary cpu clock"));
 221 }
 222 #endif
 223
 224 /*
 225  * After the clock is registered, the host will keep writing to the
 226  * registered memory location. If the guest happens to shutdown, this memory
 227  * won't be valid. In cases like kexec, in which you install a new kernel, this
 228  * means a random memory location will be kept being written. So before any
 229  * kind of shutdown from our side, we unregister the clock by writing anything
 230  * that does not have the 'enable' bit set in the msr
 231  */
 232 #ifdef CONFIG_KEXEC_CORE
 233 static void kvm_crash_shutdown(struct pt_regs *regs)
 234 {
 235         native_write_msr(msr_kvm_system_time, 0, 0);
 236         kvm_disable_steal_time();
 237         native_machine_crash_shutdown(regs);
 238 }
 239 #endif
 240
 241 static void kvm_shutdown(void)
 242 {
 243         native_write_msr(msr_kvm_system_time, 0, 0);
 244         kvm_disable_steal_time();
 245         native_machine_shutdown();
 246 }
 247
 248 void __init kvmclock_init(void)
 249 {
 250         struct pvclock_vcpu_time_info *vcpu_time;
 251         unsigned long mem;
 252         int size, cpu;
 253         u8 flags;
 254
 255         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 256
 257         if (!kvm_para_available())
 258                 return;
 259
 260         if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
 261                 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
 262                 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
 263         } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
 264                 return;
 265
 266         printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
 267                 msr_kvm_system_time, msr_kvm_wall_clock);
 268
 269         mem = memblock_alloc(size, PAGE_SIZE);
 270         if (!mem)
 271                 return;
 272         hv_clock = __va(mem);
 273         memset(hv_clock, 0, size);
 274
 275         if (kvm_register_clock("primary cpu clock")) {
 276                 hv_clock = NULL;
 277                 memblock_free(mem, size);
 278                 return;
 279         }
 280
 281         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 282                 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 283
 284         cpu = get_cpu();
 285         vcpu_time = &hv_clock[cpu].pvti;
 286         flags = pvclock_read_flags(vcpu_time);
 287
 288         kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
 289         put_cpu();
 290
 291         x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 292         x86_platform.get_wallclock = kvm_get_wallclock;
 293         x86_platform.set_wallclock = kvm_set_wallclock;
 294 #ifdef CONFIG_X86_LOCAL_APIC
 295         x86_cpuinit.early_percpu_clock_init =
 296                 kvm_setup_secondary_clock;
 297 #endif
 298         x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
 299         x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
 300         machine_ops.shutdown  = kvm_shutdown;
 301 #ifdef CONFIG_KEXEC_CORE
 302         machine_ops.crash_shutdown  = kvm_crash_shutdown;
 303 #endif
 304         kvm_get_preset_lpj();
 305         clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
 306         pv_info.name = "KVM";
 307 }
 308
 309 int __init kvm_setup_vsyscall_timeinfo(void)
 310 {
 311 #ifdef CONFIG_X86_64
 312         int cpu;
 313         u8 flags;
 314         struct pvclock_vcpu_time_info *vcpu_time;
 315         unsigned int size;
 316
 317         if (!hv_clock)
 318                 return 0;
 319
 320         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 321
 322         cpu = get_cpu();
 323
 324         vcpu_time = &hv_clock[cpu].pvti;
 325         flags = pvclock_read_flags(vcpu_time);
 326
 327         if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
 328                 put_cpu();
 329                 return 1;
 330         }
 331
 332         put_cpu();
 333
 334         kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
 335 #endif
 336         return 0;
 337 }