Commit | Line | Data |
---|---|---|
a75244c3 PO |
1 | /* |
2 | * Copyright (C) 2009 Intel Corporation. | |
3 | * Author: Patrick Ohly <patrick.ohly@intel.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
16 | * along with this program; if not, write to the Free Software | |
17 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
18 | */ | |
19 | ||
20 | #include <linux/timecompare.h> | |
21 | #include <linux/module.h> | |
5a0e3ad6 | 22 | #include <linux/slab.h> |
a75244c3 PO |
23 | #include <linux/math64.h> |
24 | ||
25 | /* | |
26 | * fixed point arithmetic scale factor for skew | |
27 | * | |
28 | * Usually one would measure skew in ppb (parts per billion, 1e9), but | |
29 | * using a factor of 2 simplifies the math. | |
30 | */ | |
31 | #define TIMECOMPARE_SKEW_RESOLUTION (((s64)1)<<30) | |
32 | ||
33 | ktime_t timecompare_transform(struct timecompare *sync, | |
34 | u64 source_tstamp) | |
35 | { | |
36 | u64 nsec; | |
37 | ||
38 | nsec = source_tstamp + sync->offset; | |
39 | nsec += (s64)(source_tstamp - sync->last_update) * sync->skew / | |
40 | TIMECOMPARE_SKEW_RESOLUTION; | |
41 | ||
42 | return ns_to_ktime(nsec); | |
43 | } | |
3586e0a9 | 44 | EXPORT_SYMBOL_GPL(timecompare_transform); |
a75244c3 PO |
45 | |
46 | int timecompare_offset(struct timecompare *sync, | |
47 | s64 *offset, | |
48 | u64 *source_tstamp) | |
49 | { | |
50 | u64 start_source = 0, end_source = 0; | |
51 | struct { | |
52 | s64 offset; | |
53 | s64 duration_target; | |
54 | } buffer[10], sample, *samples; | |
55 | int counter = 0, i; | |
56 | int used; | |
57 | int index; | |
58 | int num_samples = sync->num_samples; | |
59 | ||
60 | if (num_samples > sizeof(buffer)/sizeof(buffer[0])) { | |
61 | samples = kmalloc(sizeof(*samples) * num_samples, GFP_ATOMIC); | |
62 | if (!samples) { | |
63 | samples = buffer; | |
64 | num_samples = sizeof(buffer)/sizeof(buffer[0]); | |
65 | } | |
66 | } else { | |
67 | samples = buffer; | |
68 | } | |
69 | ||
70 | /* run until we have enough valid samples, but do not try forever */ | |
71 | i = 0; | |
72 | counter = 0; | |
73 | while (1) { | |
74 | u64 ts; | |
75 | ktime_t start, end; | |
76 | ||
77 | start = sync->target(); | |
78 | ts = timecounter_read(sync->source); | |
79 | end = sync->target(); | |
80 | ||
81 | if (!i) | |
82 | start_source = ts; | |
83 | ||
84 | /* ignore negative durations */ | |
85 | sample.duration_target = ktime_to_ns(ktime_sub(end, start)); | |
86 | if (sample.duration_target >= 0) { | |
87 | /* | |
88 | * assume symetric delay to and from source: | |
89 | * average target time corresponds to measured | |
90 | * source time | |
91 | */ | |
92 | sample.offset = | |
f065f41f | 93 | (ktime_to_ns(end) + ktime_to_ns(start)) / 2 - |
a75244c3 PO |
94 | ts; |
95 | ||
96 | /* simple insertion sort based on duration */ | |
97 | index = counter - 1; | |
98 | while (index >= 0) { | |
99 | if (samples[index].duration_target < | |
100 | sample.duration_target) | |
101 | break; | |
102 | samples[index + 1] = samples[index]; | |
103 | index--; | |
104 | } | |
105 | samples[index + 1] = sample; | |
106 | counter++; | |
107 | } | |
108 | ||
109 | i++; | |
110 | if (counter >= num_samples || i >= 100000) { | |
111 | end_source = ts; | |
112 | break; | |
113 | } | |
114 | } | |
115 | ||
116 | *source_tstamp = (end_source + start_source) / 2; | |
117 | ||
118 | /* remove outliers by only using 75% of the samples */ | |
119 | used = counter * 3 / 4; | |
120 | if (!used) | |
121 | used = counter; | |
122 | if (used) { | |
123 | /* calculate average */ | |
124 | s64 off = 0; | |
125 | for (index = 0; index < used; index++) | |
126 | off += samples[index].offset; | |
127 | *offset = div_s64(off, used); | |
128 | } | |
129 | ||
130 | if (samples && samples != buffer) | |
131 | kfree(samples); | |
132 | ||
133 | return used; | |
134 | } | |
3586e0a9 | 135 | EXPORT_SYMBOL_GPL(timecompare_offset); |
a75244c3 PO |
136 | |
137 | void __timecompare_update(struct timecompare *sync, | |
138 | u64 source_tstamp) | |
139 | { | |
140 | s64 offset; | |
141 | u64 average_time; | |
142 | ||
143 | if (!timecompare_offset(sync, &offset, &average_time)) | |
144 | return; | |
145 | ||
146 | if (!sync->last_update) { | |
147 | sync->last_update = average_time; | |
148 | sync->offset = offset; | |
149 | sync->skew = 0; | |
150 | } else { | |
151 | s64 delta_nsec = average_time - sync->last_update; | |
152 | ||
153 | /* avoid division by negative or small deltas */ | |
154 | if (delta_nsec >= 10000) { | |
155 | s64 delta_offset_nsec = offset - sync->offset; | |
156 | s64 skew; /* delta_offset_nsec * | |
157 | TIMECOMPARE_SKEW_RESOLUTION / | |
158 | delta_nsec */ | |
159 | u64 divisor; | |
160 | ||
161 | /* div_s64() is limited to 32 bit divisor */ | |
162 | skew = delta_offset_nsec * TIMECOMPARE_SKEW_RESOLUTION; | |
163 | divisor = delta_nsec; | |
164 | while (unlikely(divisor >= ((s64)1) << 32)) { | |
165 | /* divide both by 2; beware, right shift | |
166 | of negative value has undefined | |
167 | behavior and can only be used for | |
168 | the positive divisor */ | |
169 | skew = div_s64(skew, 2); | |
170 | divisor >>= 1; | |
171 | } | |
172 | skew = div_s64(skew, divisor); | |
173 | ||
174 | /* | |
175 | * Calculate new overall skew as 4/16 the | |
176 | * old value and 12/16 the new one. This is | |
177 | * a rather arbitrary tradeoff between | |
178 | * only using the latest measurement (0/16 and | |
179 | * 16/16) and even more weight on past measurements. | |
180 | */ | |
181 | #define TIMECOMPARE_NEW_SKEW_PER_16 12 | |
182 | sync->skew = | |
183 | div_s64((16 - TIMECOMPARE_NEW_SKEW_PER_16) * | |
184 | sync->skew + | |
185 | TIMECOMPARE_NEW_SKEW_PER_16 * skew, | |
186 | 16); | |
187 | sync->last_update = average_time; | |
188 | sync->offset = offset; | |
189 | } | |
190 | } | |
191 | } | |
3586e0a9 | 192 | EXPORT_SYMBOL_GPL(__timecompare_update); |