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/mem_encrypt.h>
  31 #include <asm/x86_init.h>
  32 #include <asm/reboot.h>
  33 #include <asm/kvmclock.h>
  34
  35 static int kvmclock __ro_after_init = 1;
  36 static int msr_kvm_system_time = MSR_KVM_SYSTEM_TIME;
  37 static int msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK;
  38 static u64 kvm_sched_clock_offset;
  39
  40 static int parse_no_kvmclock(char *arg)
  41 {
  42         kvmclock = 0;
  43         return 0;
  44 }
  45 early_param("no-kvmclock", parse_no_kvmclock);
  46
  47 /* The hypervisor will put information about time periodically here */
  48 static struct pvclock_vsyscall_time_info *hv_clock;
  49 static struct pvclock_wall_clock *wall_clock;
  50
  51 /*
  52  * The wallclock is the time of day when we booted. Since then, some time may
  53  * have elapsed since the hypervisor wrote the data. So we try to account for
  54  * that with system time
  55  */
  56 static void kvm_get_wallclock(struct timespec64 *now)
  57 {
  58         struct pvclock_vcpu_time_info *vcpu_time;
  59         int low, high;
  60         int cpu;
  61
  62         low = (int)slow_virt_to_phys(wall_clock);
  63         high = ((u64)slow_virt_to_phys(wall_clock) >> 32);
  64
  65         native_write_msr(msr_kvm_wall_clock, low, high);
  66
  67         cpu = get_cpu();
  68
  69         vcpu_time = &hv_clock[cpu].pvti;
  70         pvclock_read_wallclock(wall_clock, vcpu_time, now);
  71
  72         put_cpu();
  73 }
  74
  75 static int kvm_set_wallclock(const struct timespec64 *now)
  76 {
  77         return -ENODEV;
  78 }
  79
  80 static u64 kvm_clock_read(void)
  81 {
  82         struct pvclock_vcpu_time_info *src;
  83         u64 ret;
  84         int cpu;
  85
  86         preempt_disable_notrace();
  87         cpu = smp_processor_id();
  88         src = &hv_clock[cpu].pvti;
  89         ret = pvclock_clocksource_read(src);
  90         preempt_enable_notrace();
  91         return ret;
  92 }
  93
  94 static u64 kvm_clock_get_cycles(struct clocksource *cs)
  95 {
  96         return kvm_clock_read();
  97 }
  98
  99 static u64 kvm_sched_clock_read(void)
 100 {
 101         return kvm_clock_read() - kvm_sched_clock_offset;
 102 }
 103
 104 static inline void kvm_sched_clock_init(bool stable)
 105 {
 106         if (!stable) {
 107                 pv_time_ops.sched_clock = kvm_clock_read;
 108                 clear_sched_clock_stable();
 109                 return;
 110         }
 111
 112         kvm_sched_clock_offset = kvm_clock_read();
 113         pv_time_ops.sched_clock = kvm_sched_clock_read;
 114
 115         printk(KERN_INFO "kvm-clock: using sched offset of %llu cycles\n",
 116                         kvm_sched_clock_offset);
 117
 118         BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
 119                  sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
 120 }
 121
 122 /*
 123  * If we don't do that, there is the possibility that the guest
 124  * will calibrate under heavy load - thus, getting a lower lpj -
 125  * and execute the delays themselves without load. This is wrong,
 126  * because no delay loop can finish beforehand.
 127  * Any heuristics is subject to fail, because ultimately, a large
 128  * poll of guests can be running and trouble each other. So we preset
 129  * lpj here
 130  */
 131 static unsigned long kvm_get_tsc_khz(void)
 132 {
 133         struct pvclock_vcpu_time_info *src;
 134         int cpu;
 135         unsigned long tsc_khz;
 136
 137         cpu = get_cpu();
 138         src = &hv_clock[cpu].pvti;
 139         tsc_khz = pvclock_tsc_khz(src);
 140         put_cpu();
 141         setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
 142         return tsc_khz;
 143 }
 144
 145 static void kvm_get_preset_lpj(void)
 146 {
 147         unsigned long khz;
 148         u64 lpj;
 149
 150         khz = kvm_get_tsc_khz();
 151
 152         lpj = ((u64)khz * 1000);
 153         do_div(lpj, HZ);
 154         preset_lpj = lpj;
 155 }
 156
 157 bool kvm_check_and_clear_guest_paused(void)
 158 {
 159         bool ret = false;
 160         struct pvclock_vcpu_time_info *src;
 161         int cpu = smp_processor_id();
 162
 163         if (!hv_clock)
 164                 return ret;
 165
 166         src = &hv_clock[cpu].pvti;
 167         if ((src->flags & PVCLOCK_GUEST_STOPPED) != 0) {
 168                 src->flags &= ~PVCLOCK_GUEST_STOPPED;
 169                 pvclock_touch_watchdogs();
 170                 ret = true;
 171         }
 172
 173         return ret;
 174 }
 175
 176 struct clocksource kvm_clock = {
 177         .name = "kvm-clock",
 178         .read = kvm_clock_get_cycles,
 179         .rating = 400,
 180         .mask = CLOCKSOURCE_MASK(64),
 181         .flags = CLOCK_SOURCE_IS_CONTINUOUS,
 182 };
 183 EXPORT_SYMBOL_GPL(kvm_clock);
 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 static phys_addr_t __init kvm_memblock_alloc(phys_addr_t size,
 249                                              phys_addr_t align)
 250 {
 251         phys_addr_t mem;
 252
 253         mem = memblock_alloc(size, align);
 254         if (!mem)
 255                 return 0;
 256
 257         if (sev_active()) {
 258                 if (early_set_memory_decrypted((unsigned long)__va(mem), size))
 259                         goto e_free;
 260         }
 261
 262         return mem;
 263 e_free:
 264         memblock_free(mem, size);
 265         return 0;
 266 }
 267
 268 static void __init kvm_memblock_free(phys_addr_t addr, phys_addr_t size)
 269 {
 270         if (sev_active())
 271                 early_set_memory_encrypted((unsigned long)__va(addr), size);
 272
 273         memblock_free(addr, size);
 274 }
 275
 276 void __init kvmclock_init(void)
 277 {
 278         struct pvclock_vcpu_time_info *vcpu_time;
 279         unsigned long mem, mem_wall_clock;
 280         int size, cpu, wall_clock_size;
 281         u8 flags;
 282
 283         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 284
 285         if (!kvm_para_available())
 286                 return;
 287
 288         if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
 289                 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
 290                 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
 291         } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
 292                 return;
 293
 294         wall_clock_size = PAGE_ALIGN(sizeof(struct pvclock_wall_clock));
 295         mem_wall_clock = kvm_memblock_alloc(wall_clock_size, PAGE_SIZE);
 296         if (!mem_wall_clock)
 297                 return;
 298
 299         wall_clock = __va(mem_wall_clock);
 300         memset(wall_clock, 0, wall_clock_size);
 301
 302         mem = kvm_memblock_alloc(size, PAGE_SIZE);
 303         if (!mem) {
 304                 kvm_memblock_free(mem_wall_clock, wall_clock_size);
 305                 wall_clock = NULL;
 306                 return;
 307         }
 308
 309         hv_clock = __va(mem);
 310         memset(hv_clock, 0, size);
 311
 312         if (kvm_register_clock("primary cpu clock")) {
 313                 hv_clock = NULL;
 314                 kvm_memblock_free(mem, size);
 315                 kvm_memblock_free(mem_wall_clock, wall_clock_size);
 316                 wall_clock = NULL;
 317                 return;
 318         }
 319
 320         printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
 321                 msr_kvm_system_time, msr_kvm_wall_clock);
 322
 323         pvclock_set_pvti_cpu0_va(hv_clock);
 324
 325         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 326                 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 327
 328         cpu = get_cpu();
 329         vcpu_time = &hv_clock[cpu].pvti;
 330         flags = pvclock_read_flags(vcpu_time);
 331
 332         kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
 333         put_cpu();
 334
 335         x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 336         x86_platform.calibrate_cpu = kvm_get_tsc_khz;
 337         x86_platform.get_wallclock = kvm_get_wallclock;
 338         x86_platform.set_wallclock = kvm_set_wallclock;
 339 #ifdef CONFIG_X86_LOCAL_APIC
 340         x86_cpuinit.early_percpu_clock_init =
 341                 kvm_setup_secondary_clock;
 342 #endif
 343         x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
 344         x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
 345         machine_ops.shutdown  = kvm_shutdown;
 346 #ifdef CONFIG_KEXEC_CORE
 347         machine_ops.crash_shutdown  = kvm_crash_shutdown;
 348 #endif
 349         kvm_get_preset_lpj();
 350         clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
 351         pv_info.name = "KVM";
 352 }
 353
 354 int __init kvm_setup_vsyscall_timeinfo(void)
 355 {
 356 #ifdef CONFIG_X86_64
 357         int cpu;
 358         u8 flags;
 359         struct pvclock_vcpu_time_info *vcpu_time;
 360         unsigned int size;
 361
 362         if (!hv_clock)
 363                 return 0;
 364
 365         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 366
 367         cpu = get_cpu();
 368
 369         vcpu_time = &hv_clock[cpu].pvti;
 370         flags = pvclock_read_flags(vcpu_time);
 371
 372         put_cpu();
 373
 374         if (!(flags & PVCLOCK_TSC_STABLE_BIT))
 375                 return 1;
 376
 377         kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
 378 #endif
 379         return 0;
 380 }