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 #include <linux/sched/clock.h>
  29
  30 #include <asm/x86_init.h>
  31 #include <asm/reboot.h>
  32 #include <asm/kvmclock.h>
  33
  34 static int kvmclock __ro_after_init = 1;
  35 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
  36 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
  37 static u64 kvm_sched_clock_offset;
  38
  39 static int parse_no_kvmclock(char *arg)
  40 {
  41         kvmclock = 0;
  42         return 0;
  43 }
  44 early_param("no-kvmclock", parse_no_kvmclock);
  45
  46 /* The hypervisor will put information about time periodically here */
  47 static struct pvclock_vsyscall_time_info *hv_clock;
  48 static struct pvclock_wall_clock wall_clock;
  49
  50 struct pvclock_vsyscall_time_info *pvclock_pvti_cpu0_va(void)
  51 {
  52         return hv_clock;
  53 }
  54 EXPORT_SYMBOL_GPL(pvclock_pvti_cpu0_va);
  55
  56 /*
  57  * The wallclock is the time of day when we booted. Since then, some time may
  58  * have elapsed since the hypervisor wrote the data. So we try to account for
  59  * that with system time
  60  */
  61 static void kvm_get_wallclock(struct timespec *now)
  62 {
  63         struct pvclock_vcpu_time_info *vcpu_time;
  64         int low, high;
  65         int cpu;
  66
  67         low = (int)__pa_symbol(&wall_clock);
  68         high = ((u64)__pa_symbol(&wall_clock) >> 32);
  69
  70         native_write_msr(msr_kvm_wall_clock, low, high);
  71
  72         cpu = get_cpu();
  73
  74         vcpu_time = &hv_clock[cpu].pvti;
  75         pvclock_read_wallclock(&wall_clock, vcpu_time, now);
  76
  77         put_cpu();
  78 }
  79
  80 static int kvm_set_wallclock(const struct timespec *now)
  81 {
  82         return -1;
  83 }
  84
  85 static u64 kvm_clock_read(void)
  86 {
  87         struct pvclock_vcpu_time_info *src;
  88         u64 ret;
  89         int cpu;
  90
  91         preempt_disable_notrace();
  92         cpu = smp_processor_id();
  93         src = &hv_clock[cpu].pvti;
  94         ret = pvclock_clocksource_read(src);
  95         preempt_enable_notrace();
  96         return ret;
  97 }
  98
  99 static u64 kvm_clock_get_cycles(struct clocksource *cs)
 100 {
 101         return kvm_clock_read();
 102 }
 103
 104 static u64 kvm_sched_clock_read(void)
 105 {
 106         return kvm_clock_read() - kvm_sched_clock_offset;
 107 }
 108
 109 static inline void kvm_sched_clock_init(bool stable)
 110 {
 111         if (!stable) {
 112                 pv_time_ops.sched_clock = kvm_clock_read;
 113                 clear_sched_clock_stable();
 114                 return;
 115         }
 116
 117         kvm_sched_clock_offset = kvm_clock_read();
 118         pv_time_ops.sched_clock = kvm_sched_clock_read;
 119
 120         printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
 121                         kvm_sched_clock_offset);
 122
 123         BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
 124                  sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
 125 }
 126
 127 /*
 128  * If we don't do that, there is the possibility that the guest
 129  * will calibrate under heavy load - thus, getting a lower lpj -
 130  * and execute the delays themselves without load. This is wrong,
 131  * because no delay loop can finish beforehand.
 132  * Any heuristics is subject to fail, because ultimately, a large
 133  * poll of guests can be running and trouble each other. So we preset
 134  * lpj here
 135  */
 136 static unsigned long kvm_get_tsc_khz(void)
 137 {
 138         struct pvclock_vcpu_time_info *src;
 139         int cpu;
 140         unsigned long tsc_khz;
 141
 142         cpu = get_cpu();
 143         src = &hv_clock[cpu].pvti;
 144         tsc_khz = pvclock_tsc_khz(src);
 145         put_cpu();
 146         return tsc_khz;
 147 }
 148
 149 static void kvm_get_preset_lpj(void)
 150 {
 151         unsigned long khz;
 152         u64 lpj;
 153
 154         khz = kvm_get_tsc_khz();
 155
 156         lpj = ((u64)khz * 1000);
 157         do_div(lpj, HZ);
 158         preset_lpj = lpj;
 159 }
 160
 161 bool kvm_check_and_clear_guest_paused(void)
 162 {
 163         bool ret = false;
 164         struct pvclock_vcpu_time_info *src;
 165         int cpu = smp_processor_id();
 166
 167         if (!hv_clock)
 168                 return ret;
 169
 170         src = &hv_clock[cpu].pvti;
 171         if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
 172                 src->flags &= ~PVCLOCK_GUEST_STOPPED;
 173                 pvclock_touch_watchdogs();
 174                 ret = true;
 175         }
 176
 177         return ret;
 178 }
 179
 180 struct clocksource kvm_clock = {
 181         .name = "kvm-clock",
 182         .read = kvm_clock_get_cycles,
 183         .rating = 400,
 184         .mask = CLOCKSOURCE_MASK(64),
 185         .flags = CLOCK_SOURCE_IS_CONTINUOUS,
 186 };
 187 EXPORT_SYMBOL_GPL(kvm_clock);
 188
 189 int kvm_register_clock(char *txt)
 190 {
 191         int cpu = smp_processor_id();
 192         int low, high, ret;
 193         struct pvclock_vcpu_time_info *src;
 194
 195         if (!hv_clock)
 196                 return 0;
 197
 198         src = &hv_clock[cpu].pvti;
 199         low = (int)slow_virt_to_phys(src) | 1;
 200         high = ((u64)slow_virt_to_phys(src) >> 32);
 201         ret = native_write_msr_safe(msr_kvm_system_time, low, high);
 202         printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
 203                cpu, high, low, txt);
 204
 205         return ret;
 206 }
 207
 208 static void kvm_save_sched_clock_state(void)
 209 {
 210 }
 211
 212 static void kvm_restore_sched_clock_state(void)
 213 {
 214         kvm_register_clock("primary cpu clock, resume");
 215 }
 216
 217 #ifdef CONFIG_X86_LOCAL_APIC
 218 static void kvm_setup_secondary_clock(void)
 219 {
 220         /*
 221          * Now that the first cpu already had this clocksource initialized,
 222          * we shouldn't fail.
 223          */
 224         WARN_ON(kvm_register_clock("secondary cpu clock"));
 225 }
 226 #endif
 227
 228 /*
 229  * After the clock is registered, the host will keep writing to the
 230  * registered memory location. If the guest happens to shutdown, this memory
 231  * won't be valid. In cases like kexec, in which you install a new kernel, this
 232  * means a random memory location will be kept being written. So before any
 233  * kind of shutdown from our side, we unregister the clock by writing anything
 234  * that does not have the 'enable' bit set in the msr
 235  */
 236 #ifdef CONFIG_KEXEC_CORE
 237 static void kvm_crash_shutdown(struct pt_regs *regs)
 238 {
 239         native_write_msr(msr_kvm_system_time, 0, 0);
 240         kvm_disable_steal_time();
 241         native_machine_crash_shutdown(regs);
 242 }
 243 #endif
 244
 245 static void kvm_shutdown(void)
 246 {
 247         native_write_msr(msr_kvm_system_time, 0, 0);
 248         kvm_disable_steal_time();
 249         native_machine_shutdown();
 250 }
 251
 252 void __init kvmclock_init(void)
 253 {
 254         struct pvclock_vcpu_time_info *vcpu_time;
 255         unsigned long mem;
 256         int size, cpu;
 257         u8 flags;
 258
 259         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 260
 261         if (!kvm_para_available())
 262                 return;
 263
 264         if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
 265                 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
 266                 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
 267         } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
 268                 return;
 269
 270         printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
 271                 msr_kvm_system_time, msr_kvm_wall_clock);
 272
 273         mem = memblock_alloc(size, PAGE_SIZE);
 274         if (!mem)
 275                 return;
 276         hv_clock = __va(mem);
 277         memset(hv_clock, 0, size);
 278
 279         if (kvm_register_clock("primary cpu clock")) {
 280                 hv_clock = NULL;
 281                 memblock_free(mem, size);
 282                 return;
 283         }
 284
 285         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 286                 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 287
 288         cpu = get_cpu();
 289         vcpu_time = &hv_clock[cpu].pvti;
 290         flags = pvclock_read_flags(vcpu_time);
 291
 292         kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
 293         put_cpu();
 294
 295         x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 296         x86_platform.calibrate_cpu = kvm_get_tsc_khz;
 297         x86_platform.get_wallclock = kvm_get_wallclock;
 298         x86_platform.set_wallclock = kvm_set_wallclock;
 299 #ifdef CONFIG_X86_LOCAL_APIC
 300         x86_cpuinit.early_percpu_clock_init =
 301                 kvm_setup_secondary_clock;
 302 #endif
 303         x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
 304         x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
 305         machine_ops.shutdown  = kvm_shutdown;
 306 #ifdef CONFIG_KEXEC_CORE
 307         machine_ops.crash_shutdown  = kvm_crash_shutdown;
 308 #endif
 309         kvm_get_preset_lpj();
 310         clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
 311         pv_info.name = "KVM";
 312 }
 313
 314 int __init kvm_setup_vsyscall_timeinfo(void)
 315 {
 316 #ifdef CONFIG_X86_64
 317         int cpu;
 318         u8 flags;
 319         struct pvclock_vcpu_time_info *vcpu_time;
 320         unsigned int size;
 321
 322         if (!hv_clock)
 323                 return 0;
 324
 325         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 326
 327         cpu = get_cpu();
 328
 329         vcpu_time = &hv_clock[cpu].pvti;
 330         flags = pvclock_read_flags(vcpu_time);
 331
 332         if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
 333                 put_cpu();
 334                 return 1;
 335         }
 336
 337         put_cpu();
 338
 339         kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
 340 #endif
 341         return 0;
 342 }