xref: /linux/arch/x86/kernel/pvclock.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 <asm/fixmap.h>
25 #include <asm/pvclock.h>
26 
27 static u8 valid_flags __read_mostly = 0;
28 
29 void pvclock_set_flags(u8 flags)
30 {
31 	valid_flags = flags;
32 }
33 
34 unsigned long pvclock_tsc_khz(struct pvclock_vcpu_time_info *src)
35 {
36 	u64 pv_tsc_khz = 1000000ULL << 32;
37 
38 	do_div(pv_tsc_khz, src->tsc_to_system_mul);
39 	if (src->tsc_shift < 0)
40 		pv_tsc_khz <<= -src->tsc_shift;
41 	else
42 		pv_tsc_khz >>= src->tsc_shift;
43 	return pv_tsc_khz;
44 }
45 
46 static atomic64_t last_value = ATOMIC64_INIT(0);
47 
48 void pvclock_resume(void)
49 {
50 	atomic64_set(&last_value, 0);
51 }
52 
53 u8 pvclock_read_flags(struct pvclock_vcpu_time_info *src)
54 {
55 	unsigned version;
56 	cycle_t ret;
57 	u8 flags;
58 
59 	do {
60 		version = __pvclock_read_cycles(src, &ret, &flags);
61 	} while ((src->version & 1) || version != src->version);
62 
63 	return flags & valid_flags;
64 }
65 
66 cycle_t pvclock_clocksource_read(struct pvclock_vcpu_time_info *src)
67 {
68 	unsigned version;
69 	cycle_t ret;
70 	u64 last;
71 	u8 flags;
72 
73 	do {
74 		version = __pvclock_read_cycles(src, &ret, &flags);
75 	} while ((src->version & 1) || version != src->version);
76 
77 	if ((valid_flags & PVCLOCK_TSC_STABLE_BIT) &&
78 		(flags & PVCLOCK_TSC_STABLE_BIT))
79 		return ret;
80 
81 	/*
82 	 * Assumption here is that last_value, a global accumulator, always goes
83 	 * forward. If we are less than that, we should not be much smaller.
84 	 * We assume there is an error marging we're inside, and then the correction
85 	 * does not sacrifice accuracy.
86 	 *
87 	 * For reads: global may have changed between test and return,
88 	 * but this means someone else updated poked the clock at a later time.
89 	 * We just need to make sure we are not seeing a backwards event.
90 	 *
91 	 * For updates: last_value = ret is not enough, since two vcpus could be
92 	 * updating at the same time, and one of them could be slightly behind,
93 	 * making the assumption that last_value always go forward fail to hold.
94 	 */
95 	last = atomic64_read(&last_value);
96 	do {
97 		if (ret < last)
98 			return last;
99 		last = atomic64_cmpxchg(&last_value, last, ret);
100 	} while (unlikely(last != ret));
101 
102 	return ret;
103 }
104 
105 void pvclock_read_wallclock(struct pvclock_wall_clock *wall_clock,
106 			    struct pvclock_vcpu_time_info *vcpu_time,
107 			    struct timespec *ts)
108 {
109 	u32 version;
110 	u64 delta;
111 	struct timespec now;
112 
113 	/* get wallclock at system boot */
114 	do {
115 		version = wall_clock->version;
116 		rmb();		/* fetch version before time */
117 		now.tv_sec  = wall_clock->sec;
118 		now.tv_nsec = wall_clock->nsec;
119 		rmb();		/* fetch time before checking version */
120 	} while ((wall_clock->version & 1) || (version != wall_clock->version));
121 
122 	delta = pvclock_clocksource_read(vcpu_time);	/* time since system boot */
123 	delta += now.tv_sec * (u64)NSEC_PER_SEC + now.tv_nsec;
124 
125 	now.tv_nsec = do_div(delta, NSEC_PER_SEC);
126 	now.tv_sec = delta;
127 
128 	set_normalized_timespec(ts, now.tv_sec, now.tv_nsec);
129 }
130 
131 static struct pvclock_vsyscall_time_info *pvclock_vdso_info;
132 
133 static struct pvclock_vsyscall_time_info *
134 pvclock_get_vsyscall_user_time_info(int cpu)
135 {
136 	if (!pvclock_vdso_info) {
137 		BUG();
138 		return NULL;
139 	}
140 
141 	return &pvclock_vdso_info[cpu];
142 }
143 
144 struct pvclock_vcpu_time_info *pvclock_get_vsyscall_time_info(int cpu)
145 {
146 	return &pvclock_get_vsyscall_user_time_info(cpu)->pvti;
147 }
148 
149 #ifdef CONFIG_X86_64
150 static int pvclock_task_migrate(struct notifier_block *nb, unsigned long l,
151 			        void *v)
152 {
153 	struct task_migration_notifier *mn = v;
154 	struct pvclock_vsyscall_time_info *pvti;
155 
156 	pvti = pvclock_get_vsyscall_user_time_info(mn->from_cpu);
157 
158 	/* this is NULL when pvclock vsyscall is not initialized */
159 	if (unlikely(pvti == NULL))
160 		return NOTIFY_DONE;
161 
162 	pvti->migrate_count++;
163 
164 	return NOTIFY_DONE;
165 }
166 
167 static struct notifier_block pvclock_migrate = {
168 	.notifier_call = pvclock_task_migrate,
169 };
170 
171 /*
172  * Initialize the generic pvclock vsyscall state.  This will allocate
173  * a/some page(s) for the per-vcpu pvclock information, set up a
174  * fixmap mapping for the page(s)
175  */
176 
177 int __init pvclock_init_vsyscall(struct pvclock_vsyscall_time_info *i,
178 				 int size)
179 {
180 	int idx;
181 
182 	WARN_ON (size != PVCLOCK_VSYSCALL_NR_PAGES*PAGE_SIZE);
183 
184 	pvclock_vdso_info = i;
185 
186 	for (idx = 0; idx <= (PVCLOCK_FIXMAP_END-PVCLOCK_FIXMAP_BEGIN); idx++) {
187 		__set_fixmap(PVCLOCK_FIXMAP_BEGIN + idx,
188 			     __pa_symbol(i) + (idx*PAGE_SIZE),
189 			     PAGE_KERNEL_VVAR);
190 	}
191 
192 
193 	register_task_migration_notifier(&pvclock_migrate);
194 
195 	return 0;
196 }
197 #endif
198