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 /*
  51  * The wallclock is the time of day when we booted. Since then, some time may
  52  * have elapsed since the hypervisor wrote the data. So we try to account for
  53  * that with system time
  54  */
  55 static void kvm_get_wallclock(struct timespec *now)
  56 {
  57         struct pvclock_vcpu_time_info *vcpu_time;
  58         int low, high;
  59         int cpu;
  60
  61         low = (int)__pa_symbol(&wall_clock);
  62         high = ((u64)__pa_symbol(&wall_clock) >> 32);
  63
  64         native_write_msr(msr_kvm_wall_clock, low, high);
  65
  66         cpu = get_cpu();
  67
  68         vcpu_time = &hv_clock[cpu].pvti;
  69         pvclock_read_wallclock(&wall_clock, vcpu_time, now);
  70
  71         put_cpu();
  72 }
  73
  74 static int kvm_set_wallclock(const struct timespec *now)
  75 {
  76         return -ENODEV;
  77 }
  78
  79 static u64 kvm_clock_read(void)
  80 {
  81         struct pvclock_vcpu_time_info *src;
  82         u64 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 u64 kvm_clock_get_cycles(struct clocksource *cs)
  94 {
  95         return kvm_clock_read();
  96 }
  97
  98 static u64 kvm_sched_clock_read(void)
  99 {
 100         return kvm_clock_read() - kvm_sched_clock_offset;
 101 }
 102
 103 static inline void kvm_sched_clock_init(bool stable)
 104 {
 105         if (!stable) {
 106                 pv_time_ops.sched_clock = kvm_clock_read;
 107                 clear_sched_clock_stable();
 108                 return;
 109         }
 110
 111         kvm_sched_clock_offset = kvm_clock_read();
 112         pv_time_ops.sched_clock = kvm_sched_clock_read;
 113
 114         printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
 115                         kvm_sched_clock_offset);
 116
 117         BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
 118                  sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
 119 }
 120
 121 /*
 122  * If we don't do that, there is the possibility that the guest
 123  * will calibrate under heavy load - thus, getting a lower lpj -
 124  * and execute the delays themselves without load. This is wrong,
 125  * because no delay loop can finish beforehand.
 126  * Any heuristics is subject to fail, because ultimately, a large
 127  * poll of guests can be running and trouble each other. So we preset
 128  * lpj here
 129  */
 130 static unsigned long kvm_get_tsc_khz(void)
 131 {
 132         struct pvclock_vcpu_time_info *src;
 133         int cpu;
 134         unsigned long tsc_khz;
 135
 136         cpu = get_cpu();
 137         src = &hv_clock[cpu].pvti;
 138         tsc_khz = pvclock_tsc_khz(src);
 139         put_cpu();
 140         setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
 141         return tsc_khz;
 142 }
 143
 144 static void kvm_get_preset_lpj(void)
 145 {
 146         unsigned long khz;
 147         u64 lpj;
 148
 149         khz = kvm_get_tsc_khz();
 150
 151         lpj = ((u64)khz * 1000);
 152         do_div(lpj, HZ);
 153         preset_lpj = lpj;
 154 }
 155
 156 bool kvm_check_and_clear_guest_paused(void)
 157 {
 158         bool ret = false;
 159         struct pvclock_vcpu_time_info *src;
 160         int cpu = smp_processor_id();
 161
 162         if (!hv_clock)
 163                 return ret;
 164
 165         src = &hv_clock[cpu].pvti;
 166         if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
 167                 src->flags &= ~PVCLOCK_GUEST_STOPPED;
 168                 pvclock_touch_watchdogs();
 169                 ret = true;
 170         }
 171
 172         return ret;
 173 }
 174
 175 struct clocksource kvm_clock = {
 176         .name = "kvm-clock",
 177         .read = kvm_clock_get_cycles,
 178         .rating = 400,
 179         .mask = CLOCKSOURCE_MASK(64),
 180         .flags = CLOCK_SOURCE_IS_CONTINUOUS,
 181 };
 182 EXPORT_SYMBOL_GPL(kvm_clock);
 183
 184 int kvm_register_clock(char *txt)
 185 {
 186         int cpu = smp_processor_id();
 187         int low, high, ret;
 188         struct pvclock_vcpu_time_info *src;
 189
 190         if (!hv_clock)
 191                 return 0;
 192
 193         src = &hv_clock[cpu].pvti;
 194         low = (int)slow_virt_to_phys(src) | 1;
 195         high = ((u64)slow_virt_to_phys(src) >> 32);
 196         ret = native_write_msr_safe(msr_kvm_system_time, low, high);
 197         printk(KERN_INFO "kvm-clock: cpu %d, msr %x:%x, %s\n",
 198                cpu, high, low, txt);
 199
 200         return ret;
 201 }
 202
 203 static void kvm_save_sched_clock_state(void)
 204 {
 205 }
 206
 207 static void kvm_restore_sched_clock_state(void)
 208 {
 209         kvm_register_clock("primary cpu clock, resume");
 210 }
 211
 212 #ifdef CONFIG_X86_LOCAL_APIC
 213 static void kvm_setup_secondary_clock(void)
 214 {
 215         /*
 216          * Now that the first cpu already had this clocksource initialized,
 217          * we shouldn't fail.
 218          */
 219         WARN_ON(kvm_register_clock("secondary cpu clock"));
 220 }
 221 #endif
 222
 223 /*
 224  * After the clock is registered, the host will keep writing to the
 225  * registered memory location. If the guest happens to shutdown, this memory
 226  * won't be valid. In cases like kexec, in which you install a new kernel, this
 227  * means a random memory location will be kept being written. So before any
 228  * kind of shutdown from our side, we unregister the clock by writing anything
 229  * that does not have the 'enable' bit set in the msr
 230  */
 231 #ifdef CONFIG_KEXEC_CORE
 232 static void kvm_crash_shutdown(struct pt_regs *regs)
 233 {
 234         native_write_msr(msr_kvm_system_time, 0, 0);
 235         kvm_disable_steal_time();
 236         native_machine_crash_shutdown(regs);
 237 }
 238 #endif
 239
 240 static void kvm_shutdown(void)
 241 {
 242         native_write_msr(msr_kvm_system_time, 0, 0);
 243         kvm_disable_steal_time();
 244         native_machine_shutdown();
 245 }
 246
 247 void __init kvmclock_init(void)
 248 {
 249         struct pvclock_vcpu_time_info *vcpu_time;
 250         unsigned long mem;
 251         int size, cpu;
 252         u8 flags;
 253
 254         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 255
 256         if (!kvm_para_available())
 257                 return;
 258
 259         if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
 260                 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
 261                 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
 262         } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
 263                 return;
 264
 265         printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
 266                 msr_kvm_system_time, msr_kvm_wall_clock);
 267
 268         mem = memblock_alloc(size, PAGE_SIZE);
 269         if (!mem)
 270                 return;
 271         hv_clock = __va(mem);
 272         memset(hv_clock, 0, size);
 273
 274         if (kvm_register_clock("primary cpu clock")) {
 275                 hv_clock = NULL;
 276                 memblock_free(mem, size);
 277                 return;
 278         }
 279
 280         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 281                 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 282
 283         cpu = get_cpu();
 284         vcpu_time = &hv_clock[cpu].pvti;
 285         flags = pvclock_read_flags(vcpu_time);
 286
 287         kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
 288         put_cpu();
 289
 290         x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 291         x86_platform.calibrate_cpu = 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         pvclock_set_pvti_cpu0_va(hv_clock);
 333         put_cpu();
 334
 335         kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
 336 #endif
 337         return 0;
 338 }