Loading...
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
19static u8 valid_flags __read_mostly = 0;
20static struct pvclock_vsyscall_time_info *pvti_cpu0_va __read_mostly;
21
22void pvclock_set_flags(u8 flags)
23{
24 valid_flags = flags;
25}
26
27unsigned 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
39void 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
47static atomic64_t last_value = ATOMIC64_INIT(0);
48
49void pvclock_resume(void)
50{
51 atomic64_set(&last_value, 0);
52}
53
54u8 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
67static __always_inline
68u64 __pvclock_clocksource_read(struct pvclock_vcpu_time_info *src, bool dowd)
69{
70 unsigned version;
71 u64 ret;
72 u64 last;
73 u8 flags;
74
75 do {
76 version = pvclock_read_begin(src);
77 ret = __pvclock_read_cycles(src, rdtsc_ordered());
78 flags = src->flags;
79 } while (pvclock_read_retry(src, version));
80
81 if (dowd && unlikely((flags & PVCLOCK_GUEST_STOPPED) != 0)) {
82 src->flags &= ~PVCLOCK_GUEST_STOPPED;
83 pvclock_touch_watchdogs();
84 }
85
86 if ((valid_flags & PVCLOCK_TSC_STABLE_BIT) &&
87 (flags & PVCLOCK_TSC_STABLE_BIT))
88 return ret;
89
90 /*
91 * Assumption here is that last_value, a global accumulator, always goes
92 * forward. If we are less than that, we should not be much smaller.
93 * We assume there is an error margin we're inside, and then the correction
94 * does not sacrifice accuracy.
95 *
96 * For reads: global may have changed between test and return,
97 * but this means someone else updated poked the clock at a later time.
98 * We just need to make sure we are not seeing a backwards event.
99 *
100 * For updates: last_value = ret is not enough, since two vcpus could be
101 * updating at the same time, and one of them could be slightly behind,
102 * making the assumption that last_value always go forward fail to hold.
103 */
104 last = raw_atomic64_read(&last_value);
105 do {
106 if (ret <= last)
107 return last;
108 } while (!raw_atomic64_try_cmpxchg(&last_value, &last, ret));
109
110 return ret;
111}
112
113u64 pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
114{
115 return __pvclock_clocksource_read(src, true);
116}
117
118noinstr u64 pvclock_clocksource_read_nowd(struct pvclock_vcpu_time_info *src)
119{
120 return __pvclock_clocksource_read(src, false);
121}
122
123void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock,
124 struct pvclock_vcpu_time_info *vcpu_time,
125 struct timespec64 *ts)
126{
127 u32 version;
128 u64 delta;
129 struct timespec64 now;
130
131 /* get wallclock at system boot */
132 do {
133 version = wall_clock->version;
134 rmb(); /* fetch version before time */
135 /*
136 * Note: wall_clock->sec is a u32 value, so it can
137 * only store dates between 1970 and 2106. To allow
138 * times beyond that, we need to create a new hypercall
139 * interface with an extended pvclock_wall_clock structure
140 * like ARM has.
141 */
142 now.tv_sec = wall_clock->sec;
143 now.tv_nsec = wall_clock->nsec;
144 rmb(); /* fetch time before checking version */
145 } while ((wall_clock->version & 1) || (version != wall_clock->version));
146
147 delta = pvclock_clocksource_read(vcpu_time); /* time since system boot */
148 delta += now.tv_sec * NSEC_PER_SEC + now.tv_nsec;
149
150 now.tv_nsec = do_div(delta, NSEC_PER_SEC);
151 now.tv_sec = delta;
152
153 set_normalized_timespec64(ts, now.tv_sec, now.tv_nsec);
154}
155
156void pvclock_set_pvti_cpu0_va(struct pvclock_vsyscall_time_info *pvti)
157{
158 WARN_ON(vclock_was_used(VDSO_CLOCKMODE_PVCLOCK));
159 pvti_cpu0_va = pvti;
160}
161
162struct pvclock_vsyscall_time_info *pvclock_get_pvti_cpu0_va(void)
163{
164 return pvti_cpu0_va;
165}
166EXPORT_SYMBOL_GPL(pvclock_get_pvti_cpu0_va);
1/* paravirtual clock -- common code used by kvm/xen
2
3 This program is free software; you can redistribute it and/or modify
4 it under the terms of the GNU General Public License as published by
5 the Free Software Foundation; either version 2 of the License, or
6 (at your option) any later version.
7
8 This program is distributed in the hope that it will be useful,
9 but WITHOUT ANY WARRANTY; without even the implied warranty of
10 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 GNU General Public License for more details.
12
13 You should have received a copy of the GNU General Public License
14 along with this program; if not, write to the Free Software
15 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
16*/
17
18#include <linux/kernel.h>
19#include <linux/percpu.h>
20#include <asm/pvclock.h>
21
22/*
23 * These are perodically updated
24 * xen: magic shared_info page
25 * kvm: gpa registered via msr
26 * and then copied here.
27 */
28struct pvclock_shadow_time {
29 u64 tsc_timestamp; /* TSC at last update of time vals. */
30 u64 system_timestamp; /* Time, in nanosecs, since boot. */
31 u32 tsc_to_nsec_mul;
32 int tsc_shift;
33 u32 version;
34 u8 flags;
35};
36
37static u8 valid_flags __read_mostly = 0;
38
39void pvclock_set_flags(u8 flags)
40{
41 valid_flags = flags;
42}
43
44static u64 pvclock_get_nsec_offset(struct pvclock_shadow_time *shadow)
45{
46 u64 delta = native_read_tsc() - shadow->tsc_timestamp;
47 return pvclock_scale_delta(delta, shadow->tsc_to_nsec_mul,
48 shadow->tsc_shift);
49}
50
51/*
52 * Reads a consistent set of time-base values from hypervisor,
53 * into a shadow data area.
54 */
55static unsigned pvclock_get_time_values(struct pvclock_shadow_time *dst,
56 struct pvclock_vcpu_time_info *src)
57{
58 do {
59 dst->version = src->version;
60 rmb(); /* fetch version before data */
61 dst->tsc_timestamp = src->tsc_timestamp;
62 dst->system_timestamp = src->system_time;
63 dst->tsc_to_nsec_mul = src->tsc_to_system_mul;
64 dst->tsc_shift = src->tsc_shift;
65 dst->flags = src->flags;
66 rmb(); /* test version after fetching data */
67 } while ((src->version & 1) || (dst->version != src->version));
68
69 return dst->version;
70}
71
72unsigned long pvclock_tsc_khz(struct pvclock_vcpu_time_info *src)
73{
74 u64 pv_tsc_khz = 1000000ULL << 32;
75
76 do_div(pv_tsc_khz, src->tsc_to_system_mul);
77 if (src->tsc_shift < 0)
78 pv_tsc_khz <<= -src->tsc_shift;
79 else
80 pv_tsc_khz >>= src->tsc_shift;
81 return pv_tsc_khz;
82}
83
84static atomic64_t last_value = ATOMIC64_INIT(0);
85
86void pvclock_resume(void)
87{
88 atomic64_set(&last_value, 0);
89}
90
91cycle_t pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
92{
93 struct pvclock_shadow_time shadow;
94 unsigned version;
95 cycle_t ret, offset;
96 u64 last;
97
98 do {
99 version = pvclock_get_time_values(&shadow, src);
100 barrier();
101 offset = pvclock_get_nsec_offset(&shadow);
102 ret = shadow.system_timestamp + offset;
103 barrier();
104 } while (version != src->version);
105
106 if ((valid_flags & PVCLOCK_TSC_STABLE_BIT) &&
107 (shadow.flags & PVCLOCK_TSC_STABLE_BIT))
108 return ret;
109
110 /*
111 * Assumption here is that last_value, a global accumulator, always goes
112 * forward. If we are less than that, we should not be much smaller.
113 * We assume there is an error marging we're inside, and then the correction
114 * does not sacrifice accuracy.
115 *
116 * For reads: global may have changed between test and return,
117 * but this means someone else updated poked the clock at a later time.
118 * We just need to make sure we are not seeing a backwards event.
119 *
120 * For updates: last_value = ret is not enough, since two vcpus could be
121 * updating at the same time, and one of them could be slightly behind,
122 * making the assumption that last_value always go forward fail to hold.
123 */
124 last = atomic64_read(&last_value);
125 do {
126 if (ret < last)
127 return last;
128 last = atomic64_cmpxchg(&last_value, last, ret);
129 } while (unlikely(last != ret));
130
131 return ret;
132}
133
134void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock,
135 struct pvclock_vcpu_time_info *vcpu_time,
136 struct timespec *ts)
137{
138 u32 version;
139 u64 delta;
140 struct timespec now;
141
142 /* get wallclock at system boot */
143 do {
144 version = wall_clock->version;
145 rmb(); /* fetch version before time */
146 now.tv_sec = wall_clock->sec;
147 now.tv_nsec = wall_clock->nsec;
148 rmb(); /* fetch time before checking version */
149 } while ((wall_clock->version & 1) || (version != wall_clock->version));
150
151 delta = pvclock_clocksource_read(vcpu_time); /* time since system boot */
152 delta += now.tv_sec * (u64)NSEC_PER_SEC + now.tv_nsec;
153
154 now.tv_nsec = do_div(delta, NSEC_PER_SEC);
155 now.tv_sec = delta;
156
157 set_normalized_timespec(ts, now.tv_sec, now.tv_nsec);
158}