xref: /linux/arch/x86/kernel/pvclock.c (revision 856e7c4b619af622d56b3b454f7bec32a170ac99)
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 <linux/notifier.h>
21 #include <linux/sched.h>
22 #include <linux/gfp.h>
23 #include <linux/bootmem.h>
24 #include <linux/nmi.h>
25 
26 #include <asm/fixmap.h>
27 #include <asm/pvclock.h>
28 #include <asm/vgtod.h>
29 
30 static u8 valid_flags __read_mostly = 0;
31 static struct pvclock_vsyscall_time_info *pvti_cpu0_va __read_mostly;
32 
33 void pvclock_set_flags(u8 flags)
34 {
35 	valid_flags = flags;
36 }
37 
38 unsigned long pvclock_tsc_khz(struct pvclock_vcpu_time_info *src)
39 {
40 	u64 pv_tsc_khz = 1000000ULL << 32;
41 
42 	do_div(pv_tsc_khz, src->tsc_to_system_mul);
43 	if (src->tsc_shift < 0)
44 		pv_tsc_khz <<= -src->tsc_shift;
45 	else
46 		pv_tsc_khz >>= src->tsc_shift;
47 	return pv_tsc_khz;
48 }
49 
50 void pvclock_touch_watchdogs(void)
51 {
52 	touch_softlockup_watchdog_sync();
53 	clocksource_touch_watchdog();
54 	rcu_cpu_stall_reset();
55 	reset_hung_task_detector();
56 }
57 
58 static atomic64_t last_value = ATOMIC64_INIT(0);
59 
60 void pvclock_resume(void)
61 {
62 	atomic64_set(&last_value, 0);
63 }
64 
65 u8 pvclock_read_flags(struct pvclock_vcpu_time_info *src)
66 {
67 	unsigned version;
68 	u8 flags;
69 
70 	do {
71 		version = pvclock_read_begin(src);
72 		flags = src->flags;
73 	} while (pvclock_read_retry(src, version));
74 
75 	return flags & valid_flags;
76 }
77 
78 u64 pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
79 {
80 	unsigned version;
81 	u64 ret;
82 	u64 last;
83 	u8 flags;
84 
85 	do {
86 		version = pvclock_read_begin(src);
87 		ret = __pvclock_read_cycles(src, rdtsc_ordered());
88 		flags = src->flags;
89 	} while (pvclock_read_retry(src, version));
90 
91 	if (unlikely((flags & PVCLOCK_GUEST_STOPPED) != 0)) {
92 		src->flags &= ~PVCLOCK_GUEST_STOPPED;
93 		pvclock_touch_watchdogs();
94 	}
95 
96 	if ((valid_flags & PVCLOCK_TSC_STABLE_BIT) &&
97 		(flags & PVCLOCK_TSC_STABLE_BIT))
98 		return ret;
99 
100 	/*
101 	 * Assumption here is that last_value, a global accumulator, always goes
102 	 * forward. If we are less than that, we should not be much smaller.
103 	 * We assume there is an error marging we're inside, and then the correction
104 	 * does not sacrifice accuracy.
105 	 *
106 	 * For reads: global may have changed between test and return,
107 	 * but this means someone else updated poked the clock at a later time.
108 	 * We just need to make sure we are not seeing a backwards event.
109 	 *
110 	 * For updates: last_value = ret is not enough, since two vcpus could be
111 	 * updating at the same time, and one of them could be slightly behind,
112 	 * making the assumption that last_value always go forward fail to hold.
113 	 */
114 	last = atomic64_read(&last_value);
115 	do {
116 		if (ret < last)
117 			return last;
118 		last = atomic64_cmpxchg(&last_value, last, ret);
119 	} while (unlikely(last != ret));
120 
121 	return ret;
122 }
123 
124 void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock,
125 			    struct pvclock_vcpu_time_info *vcpu_time,
126 			    struct timespec64 *ts)
127 {
128 	u32 version;
129 	u64 delta;
130 	struct timespec64 now;
131 
132 	/* get wallclock at system boot */
133 	do {
134 		version = wall_clock->version;
135 		rmb();		/* fetch version before time */
136 		/*
137 		 * Note: wall_clock->sec is a u32 value, so it can
138 		 * only store dates between 1970 and 2106. To allow
139 		 * times beyond that, we need to create a new hypercall
140 		 * interface with an extended pvclock_wall_clock structure
141 		 * like ARM has.
142 		 */
143 		now.tv_sec  = wall_clock->sec;
144 		now.tv_nsec = wall_clock->nsec;
145 		rmb();		/* fetch time before checking version */
146 	} while ((wall_clock->version & 1) || (version != wall_clock->version));
147 
148 	delta = pvclock_clocksource_read(vcpu_time);	/* time since system boot */
149 	delta += now.tv_sec * NSEC_PER_SEC + now.tv_nsec;
150 
151 	now.tv_nsec = do_div(delta, NSEC_PER_SEC);
152 	now.tv_sec = delta;
153 
154 	set_normalized_timespec64(ts, now.tv_sec, now.tv_nsec);
155 }
156 
157 void pvclock_set_pvti_cpu0_va(struct pvclock_vsyscall_time_info *pvti)
158 {
159 	WARN_ON(vclock_was_used(VCLOCK_PVCLOCK));
160 	pvti_cpu0_va = pvti;
161 }
162 
163 struct pvclock_vsyscall_time_info *pvclock_get_pvti_cpu0_va(void)
164 {
165 	return pvti_cpu0_va;
166 }
167 EXPORT_SYMBOL_GPL(pvclock_get_pvti_cpu0_va);
168