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