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  */
28 struct 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 
37 static u8 valid_flags __read_mostly = 0;
38 
pvclock_set_flags(u8 flags)39 void pvclock_set_flags(u8 flags)
40 {
41 	valid_flags = flags;
42 }
43 
pvclock_get_nsec_offset(struct pvclock_shadow_time * shadow)44 static 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  */
pvclock_get_time_values(struct pvclock_shadow_time * dst,struct pvclock_vcpu_time_info * src)55 static 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 
pvclock_tsc_khz(struct pvclock_vcpu_time_info * src)72 unsigned 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 
84 static atomic64_t last_value = ATOMIC64_INIT(0);
85 
pvclock_resume(void)86 void pvclock_resume(void)
87 {
88 	atomic64_set(&last_value, 0);
89 }
90 
pvclock_clocksource_read(struct pvclock_vcpu_time_info * src)91 cycle_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 
pvclock_read_wallclock(struct pvclock_wall_clock * wall_clock,struct pvclock_vcpu_time_info * vcpu_time,struct timespec * ts)134 void 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 }
159