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/cpuhotplug.h>
  27 #include <linux/sched.h>
  28 #include <linux/sched/clock.h>
  29 #include <linux/mm.h>
  30 #include <linux/slab.h>
  31 #include <linux/set_memory.h>
  32
  33 #include <asm/hypervisor.h>
  34 #include <asm/mem_encrypt.h>
  35 #include <asm/x86_init.h>
  36 #include <asm/reboot.h>
  37 #include <asm/kvmclock.h>
  38
  39 static int kvmclock __initdata = 1;
  40 static int kvmclock_vsyscall __initdata = 1;
  41 static int msr_kvm_system_time __ro_after_init = MSR_KVM_SYSTEM_TIME;
  42 static int msr_kvm_wall_clock __ro_after_init = MSR_KVM_WALL_CLOCK;
  43 static u64 kvm_sched_clock_offset __ro_after_init;
  44
  45 static int __init parse_no_kvmclock(char *arg)
  46 {
  47         kvmclock = 0;
  48         return 0;
  49 }
  50 early_param("no-kvmclock", parse_no_kvmclock);
  51
  52 static int __init parse_no_kvmclock_vsyscall(char *arg)
  53 {
  54         kvmclock_vsyscall = 0;
  55         return 0;
  56 }
  57 early_param("no-kvmclock-vsyscall", parse_no_kvmclock_vsyscall);
  58
  59 /* Aligned to page sizes to match whats mapped via vsyscalls to userspace */
  60 #define HV_CLOCK_SIZE   (sizeof(struct pvclock_vsyscall_time_info) * NR_CPUS)
  61 #define HVC_BOOT_ARRAY_SIZE \
  62         (PAGE_SIZE / sizeof(struct pvclock_vsyscall_time_info))
  63
  64 static struct pvclock_vsyscall_time_info
  65                         hv_clock_boot[HVC_BOOT_ARRAY_SIZE] __bss_decrypted __aligned(PAGE_SIZE);
  66 static struct pvclock_wall_clock wall_clock __bss_decrypted;
  67 static DEFINE_PER_CPU(struct pvclock_vsyscall_time_info *, hv_clock_per_cpu);
  68 static struct pvclock_vsyscall_time_info *hvclock_mem;
  69
  70 static inline struct pvclock_vcpu_time_info *this_cpu_pvti(void)
  71 {
  72         return &this_cpu_read(hv_clock_per_cpu)->pvti;
  73 }
  74
  75 static inline struct pvclock_vsyscall_time_info *this_cpu_hvclock(void)
  76 {
  77         return this_cpu_read(hv_clock_per_cpu);
  78 }
  79
  80 /*
  81  * The wallclock is the time of day when we booted. Since then, some time may
  82  * have elapsed since the hypervisor wrote the data. So we try to account for
  83  * that with system time
  84  */
  85 static void kvm_get_wallclock(struct timespec64 *now)
  86 {
  87         wrmsrl(msr_kvm_wall_clock, slow_virt_to_phys(&wall_clock));
  88         preempt_disable();
  89         pvclock_read_wallclock(&wall_clock, this_cpu_pvti(), now);
  90         preempt_enable();
  91 }
  92
  93 static int kvm_set_wallclock(const struct timespec64 *now)
  94 {
  95         return -ENODEV;
  96 }
  97
  98 static u64 kvm_clock_read(void)
  99 {
 100         u64 ret;
 101
 102         preempt_disable_notrace();
 103         ret = pvclock_clocksource_read(this_cpu_pvti());
 104         preempt_enable_notrace();
 105         return ret;
 106 }
 107
 108 static u64 kvm_clock_get_cycles(struct clocksource *cs)
 109 {
 110         return kvm_clock_read();
 111 }
 112
 113 static u64 kvm_sched_clock_read(void)
 114 {
 115         return kvm_clock_read() - kvm_sched_clock_offset;
 116 }
 117
 118 static inline void kvm_sched_clock_init(bool stable)
 119 {
 120         if (!stable) {
 121                 pv_ops.time.sched_clock = kvm_clock_read;
 122                 clear_sched_clock_stable();
 123                 return;
 124         }
 125
 126         kvm_sched_clock_offset = kvm_clock_read();
 127         pv_ops.time.sched_clock = kvm_sched_clock_read;
 128
 129         pr_info("kvm-clock: using sched offset of %llu cycles",
 130                 kvm_sched_clock_offset);
 131
 132         BUILD_BUG_ON(sizeof(kvm_sched_clock_offset) >
 133                 sizeof(((struct pvclock_vcpu_time_info *)NULL)->system_time));
 134 }
 135
 136 /*
 137  * If we don't do that, there is the possibility that the guest
 138  * will calibrate under heavy load - thus, getting a lower lpj -
 139  * and execute the delays themselves without load. This is wrong,
 140  * because no delay loop can finish beforehand.
 141  * Any heuristics is subject to fail, because ultimately, a large
 142  * poll of guests can be running and trouble each other. So we preset
 143  * lpj here
 144  */
 145 static unsigned long kvm_get_tsc_khz(void)
 146 {
 147         setup_force_cpu_cap(X86_FEATURE_TSC_KNOWN_FREQ);
 148         return pvclock_tsc_khz(this_cpu_pvti());
 149 }
 150
 151 static void __init kvm_get_preset_lpj(void)
 152 {
 153         unsigned long khz;
 154         u64 lpj;
 155
 156         khz = kvm_get_tsc_khz();
 157
 158         lpj = ((u64)khz * 1000);
 159         do_div(lpj, HZ);
 160         preset_lpj = lpj;
 161 }
 162
 163 bool kvm_check_and_clear_guest_paused(void)
 164 {
 165         struct pvclock_vsyscall_time_info *src = this_cpu_hvclock();
 166         bool ret = false;
 167
 168         if (!src)
 169                 return ret;
 170
 171         if ((src->pvti.flags & PVCLOCK_GUEST_STOPPED) != 0) {
 172                 src->pvti.flags &= ~PVCLOCK_GUEST_STOPPED;
 173                 pvclock_touch_watchdogs();
 174                 ret = true;
 175         }
 176         return ret;
 177 }
 178
 179 struct clocksource kvm_clock = {
 180         .name   = "kvm-clock",
 181         .read   = kvm_clock_get_cycles,
 182         .rating = 400,
 183         .mask   = CLOCKSOURCE_MASK(64),
 184         .flags  = CLOCK_SOURCE_IS_CONTINUOUS,
 185 };
 186 EXPORT_SYMBOL_GPL(kvm_clock);
 187
 188 static void kvm_register_clock(char *txt)
 189 {
 190         struct pvclock_vsyscall_time_info *src = this_cpu_hvclock();
 191         u64 pa;
 192
 193         if (!src)
 194                 return;
 195
 196         pa = slow_virt_to_phys(&src->pvti) | 0x01ULL;
 197         wrmsrl(msr_kvm_system_time, pa);
 198         pr_info("kvm-clock: cpu %d, msr %llx, %s", smp_processor_id(), pa, txt);
 199 }
 200
 201 static void kvm_save_sched_clock_state(void)
 202 {
 203 }
 204
 205 static void kvm_restore_sched_clock_state(void)
 206 {
 207         kvm_register_clock("primary cpu clock, resume");
 208 }
 209
 210 #ifdef CONFIG_X86_LOCAL_APIC
 211 static void kvm_setup_secondary_clock(void)
 212 {
 213         kvm_register_clock("secondary cpu clock");
 214 }
 215 #endif
 216
 217 /*
 218  * After the clock is registered, the host will keep writing to the
 219  * registered memory location. If the guest happens to shutdown, this memory
 220  * won't be valid. In cases like kexec, in which you install a new kernel, this
 221  * means a random memory location will be kept being written. So before any
 222  * kind of shutdown from our side, we unregister the clock by writing anything
 223  * that does not have the 'enable' bit set in the msr
 224  */
 225 #ifdef CONFIG_KEXEC_CORE
 226 static void kvm_crash_shutdown(struct pt_regs *regs)
 227 {
 228         native_write_msr(msr_kvm_system_time, 0, 0);
 229         kvm_disable_steal_time();
 230         native_machine_crash_shutdown(regs);
 231 }
 232 #endif
 233
 234 static void kvm_shutdown(void)
 235 {
 236         native_write_msr(msr_kvm_system_time, 0, 0);
 237         kvm_disable_steal_time();
 238         native_machine_shutdown();
 239 }
 240
 241 static void __init kvmclock_init_mem(void)
 242 {
 243         unsigned long ncpus;
 244         unsigned int order;
 245         struct page *p;
 246         int r;
 247
 248         if (HVC_BOOT_ARRAY_SIZE >= num_possible_cpus())
 249                 return;
 250
 251         ncpus = num_possible_cpus() - HVC_BOOT_ARRAY_SIZE;
 252         order = get_order(ncpus * sizeof(*hvclock_mem));
 253
 254         p = alloc_pages(GFP_KERNEL, order);
 255         if (!p) {
 256                 pr_warn("%s: failed to alloc %d pages", __func__, (1U << order));
 257                 return;
 258         }
 259
 260         hvclock_mem = page_address(p);
 261
 262         /*
 263          * hvclock is shared between the guest and the hypervisor, must
 264          * be mapped decrypted.
 265          */
 266         if (sev_active()) {
 267                 r = set_memory_decrypted((unsigned long) hvclock_mem,
 268                                          1UL << order);
 269                 if (r) {
 270                         __free_pages(p, order);
 271                         hvclock_mem = NULL;
 272                         pr_warn("kvmclock: set_memory_decrypted() failed. Disabling\n");
 273                         return;
 274                 }
 275         }
 276
 277         memset(hvclock_mem, 0, PAGE_SIZE << order);
 278 }
 279
 280 static int __init kvm_setup_vsyscall_timeinfo(void)
 281 {
 282 #ifdef CONFIG_X86_64
 283         u8 flags;
 284
 285         if (!per_cpu(hv_clock_per_cpu, 0) || !kvmclock_vsyscall)
 286                 return 0;
 287
 288         flags = pvclock_read_flags(&hv_clock_boot[0].pvti);
 289         if (!(flags & PVCLOCK_TSC_STABLE_BIT))
 290                 return 0;
 291
 292         kvm_clock.archdata.vclock_mode = VCLOCK_PVCLOCK;
 293 #endif
 294
 295         kvmclock_init_mem();
 296
 297         return 0;
 298 }
 299 early_initcall(kvm_setup_vsyscall_timeinfo);
 300
 301 static int kvmclock_setup_percpu(unsigned int cpu)
 302 {
 303         struct pvclock_vsyscall_time_info *p = per_cpu(hv_clock_per_cpu, cpu);
 304
 305         /*
 306          * The per cpu area setup replicates CPU0 data to all cpu
 307          * pointers. So carefully check. CPU0 has been set up in init
 308          * already.
 309          */
 310         if (!cpu || (p && p != per_cpu(hv_clock_per_cpu, 0)))
 311                 return 0;
 312
 313         /* Use the static page for the first CPUs, allocate otherwise */
 314         if (cpu < HVC_BOOT_ARRAY_SIZE)
 315                 p = &hv_clock_boot[cpu];
 316         else if (hvclock_mem)
 317                 p = hvclock_mem + cpu - HVC_BOOT_ARRAY_SIZE;
 318         else
 319                 return -ENOMEM;
 320
 321         per_cpu(hv_clock_per_cpu, cpu) = p;
 322         return p ? 0 : -ENOMEM;
 323 }
 324
 325 void __init kvmclock_init(void)
 326 {
 327         u8 flags;
 328
 329         if (!kvm_para_available() || !kvmclock)
 330                 return;
 331
 332         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE2)) {
 333                 msr_kvm_system_time = MSR_KVM_SYSTEM_TIME_NEW;
 334                 msr_kvm_wall_clock = MSR_KVM_WALL_CLOCK_NEW;
 335         } else if (!kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE)) {
 336                 return;
 337         }
 338
 339         if (cpuhp_setup_state(CPUHP_BP_PREPARE_DYN, "kvmclock:setup_percpu",
 340                               kvmclock_setup_percpu, NULL) < 0) {
 341                 return;
 342         }
 343
 344         pr_info("kvm-clock: Using msrs %x and %x",
 345                 msr_kvm_system_time, msr_kvm_wall_clock);
 346
 347         this_cpu_write(hv_clock_per_cpu, &hv_clock_boot[0]);
 348         kvm_register_clock("primary cpu clock");
 349         pvclock_set_pvti_cpu0_va(hv_clock_boot);
 350
 351         if (kvm_para_has_feature(KVM_FEATURE_CLOCKSOURCE_STABLE_BIT))
 352                 pvclock_set_flags(PVCLOCK_TSC_STABLE_BIT);
 353
 354         flags = pvclock_read_flags(&hv_clock_boot[0].pvti);
 355         kvm_sched_clock_init(flags & PVCLOCK_TSC_STABLE_BIT);
 356
 357         x86_platform.calibrate_tsc = kvm_get_tsc_khz;
 358         x86_platform.calibrate_cpu = kvm_get_tsc_khz;
 359         x86_platform.get_wallclock = kvm_get_wallclock;
 360         x86_platform.set_wallclock = kvm_set_wallclock;
 361 #ifdef CONFIG_X86_LOCAL_APIC
 362         x86_cpuinit.early_percpu_clock_init = kvm_setup_secondary_clock;
 363 #endif
 364         x86_platform.save_sched_clock_state = kvm_save_sched_clock_state;
 365         x86_platform.restore_sched_clock_state = kvm_restore_sched_clock_state;
 366         machine_ops.shutdown  = kvm_shutdown;
 367 #ifdef CONFIG_KEXEC_CORE
 368         machine_ops.crash_shutdown  = kvm_crash_shutdown;
 369 #endif
 370         kvm_get_preset_lpj();
 371         clocksource_register_hz(&kvm_clock, NSEC_PER_SEC);
 372         pv_info.name = "KVM";
 373 }