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 timespec *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 timespec *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         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 static phys_addr_t __init kvm_memblock_alloc(phys_addr_t size,
 248                                              phys_addr_t align)
 249 {
 250         phys_addr_t mem;
 251
 252         mem = memblock_alloc(size, align);
 253         if (!mem)
 254                 return 0;
 255
 256         if (sev_active()) {
 257                 if (early_set_memory_decrypted((unsigned long)__va(mem), size))
 258                         goto e_free;
 259         }
 260
 261         return mem;
 262 e_free:
 263         memblock_free(mem, size);
 264         return 0;
 265 }
 266
 267 static void __init kvm_memblock_free(phys_addr_t addr, phys_addr_t size)
 268 {
 269         if (sev_active())
 270                 early_set_memory_encrypted((unsigned long)__va(addr), size);
 271
 272         memblock_free(addr, size);
 273 }
 274
 275 void __init kvmclock_init(void)
 276 {
 277         struct pvclock_vcpu_time_info *vcpu_time;
 278         unsigned long mem, mem_wall_clock;
 279         int size, cpu, wall_clock_size;
 280         u8 flags;
 281
 282         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 283
 284         if (!kvm_para_available())
 285                 return;
 286
 287         if (kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
 288                 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
 289                 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
 290         } else if (!(kvmclock && kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)))
 291                 return;
 292
 293         wall_clock_size = PAGE_ALIGN(sizeof(struct pvclock_wall_clock));
 294         mem_wall_clock = kvm_memblock_alloc(wall_clock_size, PAGE_SIZE);
 295         if (!mem_wall_clock)
 296                 return;
 297
 298         wall_clock = __va(mem_wall_clock);
 299         memset(wall_clock, 0, wall_clock_size);
 300
 301         mem = kvm_memblock_alloc(size, PAGE_SIZE);
 302         if (!mem) {
 303                 kvm_memblock_free(mem_wall_clock, wall_clock_size);
 304                 wall_clock = NULL;
 305                 return;
 306         }
 307
 308         hv_clock = __va(mem);
 309         memset(hv_clock, 0, size);
 310
 311         if (kvm_register_clock("primary cpu clock")) {
 312                 hv_clock = NULL;
 313                 kvm_memblock_free(mem, size);
 314                 kvm_memblock_free(mem_wall_clock, wall_clock_size);
 315                 wall_clock = NULL;
 316                 return;
 317         }
 318
 319         printk(KERN_INFO "kvm-clock: Using msrs %x and %x",
 320                 msr_kvm_system_time, msr_kvm_wall_clock);
 321
 322         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 323                 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 324
 325         cpu = get_cpu();
 326         vcpu_time = &hv_clock[cpu].pvti;
 327         flags = pvclock_read_flags(vcpu_time);
 328
 329         kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
 330         put_cpu();
 331
 332         x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 333         x86_platform.calibrate_cpu = kvm_get_tsc_khz;
 334         x86_platform.get_wallclock = kvm_get_wallclock;
 335         x86_platform.set_wallclock = kvm_set_wallclock;
 336 #ifdef CONFIG_X86_LOCAL_APIC
 337         x86_cpuinit.early_percpu_clock_init =
 338                 kvm_setup_secondary_clock;
 339 #endif
 340         x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
 341         x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
 342         machine_ops.shutdown  = kvm_shutdown;
 343 #ifdef CONFIG_KEXEC_CORE
 344         machine_ops.crash_shutdown  = kvm_crash_shutdown;
 345 #endif
 346         kvm_get_preset_lpj();
 347         clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
 348         pv_info.name = "KVM";
 349 }
 350
 351 int __init kvm_setup_vsyscall_timeinfo(void)
 352 {
 353 #ifdef CONFIG_X86_64
 354         int cpu;
 355         u8 flags;
 356         struct pvclock_vcpu_time_info *vcpu_time;
 357         unsigned int size;
 358
 359         if (!hv_clock)
 360                 return 0;
 361
 362         size = PAGE_ALIGN(sizeof(struct pvclock_vsyscall_time_info)*NR_CPUS);
 363
 364         cpu = get_cpu();
 365
 366         vcpu_time = &hv_clock[cpu].pvti;
 367         flags = pvclock_read_flags(vcpu_time);
 368
 369         if (!(flags & PVCLOCK_TSC_STABLE_BIT)) {
 370                 put_cpu();
 371                 return 1;
 372         }
 373
 374         pvclock_set_pvti_cpu0_va(hv_clock);
 375         put_cpu();
 376
 377         kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
 378 #endif
 379         return 0;
 380 }