x86/kernel/pvclock.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*  paravirtual clock -- common code used by kvm/xen
0003
0004 */
0005
0006 #include <linux/clocksource.h>
0007 #include <linux/kernel.h>
0008 #include <linux/percpu.h>
0009 #include <linux/notifier.h>
0010 #include <linux/sched.h>
0011 #include <linux/gfp.h>
0012 #include <linux/memblock.h>
0013 #include <linux/nmi.h>
0014
0015 #include <asm/fixmap.h>
0016 #include <asm/pvclock.h>
0017 #include <asm/vgtod.h>
0018
0019 static u8 valid_flags __read_mostly = 0;
0020 static struct pvclock_vsyscall_time_info *pvti_cpu0_va __read_mostly;
0021
0022 void pvclock_set_flags(u8 flags)
0023 {
0024     valid_flags = flags;
0025 }
0026
0027 unsigned long pvclock_tsc_khz(struct pvclock_vcpu_time_info *src)
0028 {
0029     u64 pv_tsc_khz = 1000000ULL << 32;
0030
0031     do_div(pv_tsc_khz, src->tsc_to_system_mul);
0032     if (src->tsc_shift < 0)
0033         pv_tsc_khz <<= -src->tsc_shift;
0034     else
0035         pv_tsc_khz >>= src->tsc_shift;
0036     return pv_tsc_khz;
0037 }
0038
0039 void pvclock_touch_watchdogs(void)
0040 {
0041     touch_softlockup_watchdog_sync();
0042     clocksource_touch_watchdog();
0043     rcu_cpu_stall_reset();
0044     reset_hung_task_detector();
0045 }
0046
0047 static atomic64_t last_value = ATOMIC64_INIT(0);
0048
0049 void pvclock_resume(void)
0050 {
0051     atomic64_set(&last_value, 0);
0052 }
0053
0054 u8 pvclock_read_flags(struct pvclock_vcpu_time_info *src)
0055 {
0056     unsigned version;
0057     u8 flags;
0058
0059     do {
0060         version = pvclock_read_begin(src);
0061         flags = src->flags;
0062     } while (pvclock_read_retry(src, version));
0063
0064     return flags & valid_flags;
0065 }
0066
0067 u64 pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
0068 {
0069     unsigned version;
0070     u64 ret;
0071     u64 last;
0072     u8 flags;
0073
0074     do {
0075         version = pvclock_read_begin(src);
0076         ret = __pvclock_read_cycles(src, rdtsc_ordered());
0077         flags = src->flags;
0078     } while (pvclock_read_retry(src, version));
0079
0080     if (unlikely((flags & PVCLOCK_GUEST_STOPPED) != 0)) {
0081         src->flags &= ~PVCLOCK_GUEST_STOPPED;
0082         pvclock_touch_watchdogs();
0083     }
0084
0085     if ((valid_flags & PVCLOCK_TSC_STABLE_BIT) &&
0086         (flags & PVCLOCK_TSC_STABLE_BIT))
0087         return ret;
0088
0089     /*
0090      * Assumption here is that last_value, a global accumulator, always goes
0091      * forward. If we are less than that, we should not be much smaller.
0092      * We assume there is an error margin we're inside, and then the correction
0093      * does not sacrifice accuracy.
0094      *
0095      * For reads: global may have changed between test and return,
0096      * but this means someone else updated poked the clock at a later time.
0097      * We just need to make sure we are not seeing a backwards event.
0098      *
0099      * For updates: last_value = ret is not enough, since two vcpus could be
0100      * updating at the same time, and one of them could be slightly behind,
0101      * making the assumption that last_value always go forward fail to hold.
0102      */
0103     last = atomic64_read(&last_value);
0104     do {
0105         if (ret < last)
0106             return last;
0107         last = atomic64_cmpxchg(&last_value, last, ret);
0108     } while (unlikely(last != ret));
0109
0110     return ret;
0111 }
0112
0113 void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock,
0114                 struct pvclock_vcpu_time_info *vcpu_time,
0115                 struct timespec64 *ts)
0116 {
0117     u32 version;
0118     u64 delta;
0119     struct timespec64 now;
0120
0121     /* get wallclock at system boot */
0122     do {
0123         version = wall_clock->version;
0124         rmb();      /* fetch version before time */
0125         /*
0126          * Note: wall_clock->sec is a u32 value, so it can
0127          * only store dates between 1970 and 2106. To allow
0128          * times beyond that, we need to create a new hypercall
0129          * interface with an extended pvclock_wall_clock structure
0130          * like ARM has.
0131          */
0132         now.tv_sec  = wall_clock->sec;
0133         now.tv_nsec = wall_clock->nsec;
0134         rmb();      /* fetch time before checking version */
0135     } while ((wall_clock->version & 1) || (version != wall_clock->version));
0136
0137     delta = pvclock_clocksource_read(vcpu_time);    /* time since system boot */
0138     delta += now.tv_sec * NSEC_PER_SEC + now.tv_nsec;
0139
0140     now.tv_nsec = do_div(delta, NSEC_PER_SEC);
0141     now.tv_sec = delta;
0142
0143     set_normalized_timespec64(ts, now.tv_sec, now.tv_nsec);
0144 }
0145
0146 void pvclock_set_pvti_cpu0_va(struct pvclock_vsyscall_time_info *pvti)
0147 {
0148     WARN_ON(vclock_was_used(VDSO_CLOCKMODE_PVCLOCK));
0149     pvti_cpu0_va = pvti;
0150 }
0151
0152 struct pvclock_vsyscall_time_info *pvclock_get_pvti_cpu0_va(void)
0153 {
0154     return pvti_cpu0_va;
0155 }
0156 EXPORT_SYMBOL_GPL(pvclock_get_pvti_cpu0_va);