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