arch/x86/kernel/pvclock.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*  paravirtual clock -- common code used by kvm/xen
   3
   4 */
   5
   6 #include <linux/clocksource.h>
   7 #include <linux/kernel.h>
   8 #include <linux/percpu.h>
   9 #include <linux/notifier.h>
  10 #include <linux/sched.h>
  11 #include <linux/gfp.h>
  12 #include <linux/memblock.h>
  13 #include <linux/nmi.h>
  14
  15 #include <asm/fixmap.h>
  16 #include <asm/pvclock.h>
  17 #include <asm/vgtod.h>
  18
  19 static u8 valid_flags __read_mostly = 0;
  20 static struct pvclock_vsyscall_time_info *pvti_cpu0_va __read_mostly;
  21
  22 void pvclock_set_flags(u8 flags)
  23 {
  24         valid_flags = flags;
  25 }
  26
  27 unsigned long pvclock_tsc_khz(struct pvclock_vcpu_time_info *src)
  28 {
  29         u64 pv_tsc_khz = 1000000ULL << 32;
  30
  31         do_div(pv_tsc_khz, src->tsc_to_system_mul);
  32         if (src->tsc_shift < 0)
  33                 pv_tsc_khz <<= -src->tsc_shift;
  34         else
  35                 pv_tsc_khz >>= src->tsc_shift;
  36         return pv_tsc_khz;
  37 }
  38
  39 void pvclock_touch_watchdogs(void)
  40 {
  41         touch_softlockup_watchdog_sync();
  42         clocksource_touch_watchdog();
  43         rcu_cpu_stall_reset();
  44         reset_hung_task_detector();
  45 }
  46
  47 static atomic64_t last_value = ATOMIC64_INIT(0);
  48
  49 void pvclock_resume(void)
  50 {
  51         atomic64_set(&last_value, 0);
  52 }
  53
  54 u8 pvclock_read_flags(struct pvclock_vcpu_time_info *src)
  55 {
  56         unsigned version;
  57         u8 flags;
  58
  59         do {
  60                 version = pvclock_read_begin(src);
  61                 flags = src->flags;
  62         } while (pvclock_read_retry(src, version));
  63
  64         return flags & valid_flags;
  65 }
  66
  67 u64 pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
  68 {
  69         unsigned version;
  70         u64 ret;
  71         u64 last;
  72         u8 flags;
  73
  74         do {
  75                 version = pvclock_read_begin(src);
  76                 ret = __pvclock_read_cycles(src, rdtsc_ordered());
  77                 flags = src->flags;
  78         } while (pvclock_read_retry(src, version));
  79
  80         if (unlikely((flags & PVCLOCK_GUEST_STOPPED) != 0)) {
  81                 src->flags &= ~PVCLOCK_GUEST_STOPPED;
  82                 pvclock_touch_watchdogs();
  83         }
  84
  85         if ((valid_flags & PVCLOCK_TSC_STABLE_BIT) &&
  86                 (flags & PVCLOCK_TSC_STABLE_BIT))
  87                 return ret;
  88
  89         /*
  90          * Assumption here is that last_value, a global accumulator, always goes
  91          * forward. If we are less than that, we should not be much smaller.
  92          * We assume there is an error marging we're inside, and then the correction
  93          * does not sacrifice accuracy.
  94          *
  95          * For reads: global may have changed between test and return,
  96          * but this means someone else updated poked the clock at a later time.
  97          * We just need to make sure we are not seeing a backwards event.
  98          *
  99          * For updates: last_value = ret is not enough, since two vcpus could be
 100          * updating at the same time, and one of them could be slightly behind,
 101          * making the assumption that last_value always go forward fail to hold.
 102          */
 103         last = atomic64_read(&last_value);
 104         do {
 105                 if (ret < last)
 106                         return last;
 107                 last = atomic64_cmpxchg(&last_value, last, ret);
 108         } while (unlikely(last != ret));
 109
 110         return ret;
 111 }
 112
 113 void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock,
 114                             struct pvclock_vcpu_time_info *vcpu_time,
 115                             struct timespec64 *ts)
 116 {
 117         u32 version;
 118         u64 delta;
 119         struct timespec64 now;
 120
 121         /* get wallclock at system boot */
 122         do {
 123                 version = wall_clock->version;
 124                 rmb();          /* fetch version before time */
 125                 /*
 126                  * Note: wall_clock->sec is a u32 value, so it can
 127                  * only store dates between 1970 and 2106. To allow
 128                  * times beyond that, we need to create a new hypercall
 129                  * interface with an extended pvclock_wall_clock structure
 130                  * like ARM has.
 131                  */
 132                 now.tv_sec  = wall_clock->sec;
 133                 now.tv_nsec = wall_clock->nsec;
 134                 rmb();          /* fetch time before checking version */
 135         } while ((wall_clock->version & 1) || (version != wall_clock->version));
 136
 137         delta = pvclock_clocksource_read(vcpu_time);    /* time since system boot */
 138         delta += now.tv_sec * NSEC_PER_SEC + now.tv_nsec;
 139
 140         now.tv_nsec = do_div(delta, NSEC_PER_SEC);
 141         now.tv_sec = delta;
 142
 143         set_normalized_timespec64(ts, now.tv_sec, now.tv_nsec);
 144 }
 145
 146 void pvclock_set_pvti_cpu0_va(struct pvclock_vsyscall_time_info *pvti)
 147 {
 148         WARN_ON(vclock_was_used(VCLOCK_PVCLOCK));
 149         pvti_cpu0_va = pvti;
 150 }
 151
 152 struct pvclock_vsyscall_time_info *pvclock_get_pvti_cpu0_va(void)
 153 {
 154         return pvti_cpu0_va;
 155 }
 156 EXPORT_SYMBOL_GPL(pvclock_get_pvti_cpu0_va);