1 ==============================================
2 Clock sources, Clock events, sched_clock() and
3 ==============================================
4
5 This document tries to briefly explain some ba
6 abstractions. It partly pertains to the driver
7 drivers/clocksource in the kernel tree, but th
8 across the kernel.
9
10 If you grep through the kernel source you will
11 specific implementations of clock sources, clo
12 architecture-specific overrides of the sched_c
13 delay timers.
14
15 To provide timekeeping for your platform, the
16 the basic timeline, whereas clock events shoot
17 on this timeline, providing facilities such as
18 sched_clock() is used for scheduling and times
19 provide an accurate delay source using hardwar
20
21
22 Clock sources
23 -------------
24
25 The purpose of the clock source is to provide
26 tells you where you are in time. For example i
27 a Linux system will eventually read the clock
28 what time it is.
29
30 Typically the clock source is a monotonic, ato
31 n bits which count from 0 to (2^n)-1 and then
32 It will ideally NEVER stop ticking as long as
33 may stop during system suspend.
34
35 The clock source shall have as high resolution
36 shall be as stable and correct as possible as
37 clock. It should not move unpredictably back a
38 cycles here and there.
39
40 It must be immune to the kind of effects that
41 the counter register is read in two phases on
42 and the higher 16 bits in a second bus cycle w
43 potentially being updated in between leading t
44 values from the counter.
45
46 When the wall-clock accuracy of the clock sour
47 are various quirks and layers in the timekeepi
48 the user-visible time to RTC clocks in the sys
49 servers using NTP, but all they do basically i
50 the clock source, which provides the fundament
51 These measures does not affect the clock sourc
52 system to the shortcomings of it.
53
54 The clock source struct shall provide means to
55 into a nanosecond value as an unsigned long lo
56 Since this operation may be invoked very often
57 mathematical sense is not desirable: instead t
58 possible to a nanosecond value using only the
59 multiply and shift, so in clocksource_cyc2ns()
60
61 ns ~= (clocksource * mult) >> shift
62
63 You will find a number of helper functions in
64 to aid in providing these mult and shift value
65 clocksource_khz2mult(), clocksource_hz2mult()
66 mult factor from a fixed shift, and clocksourc
67 clocksource_register_khz() which will help out
68 factors using the frequency of the clock sourc
69
70 For real simple clock sources accessed from a
71 there is nowadays even clocksource_mmio_init()
72 location, bit width, a parameter telling wheth
73 register counts up or down, and the timer cloc
74 necessary parameters.
75
76 Since a 32-bit counter at say 100 MHz will wra
77 seconds, the code handling the clock source wi
78 That is the reason why the clock source struct
79 member telling how many bits of the source are
80 code knows when the counter will wrap around a
81 compensation code on both sides of the wrap po
82 remains monotonic.
83
84
85 Clock events
86 ------------
87
88 Clock events are the conceptual reverse of clo
89 desired time specification value and calculate
90 hardware timer registers.
91
92 Clock events are orthogonal to clock sources.
93 and register range may be used for the clock e
94 a different thing. The hardware driving clock
95 fire interrupts, so as to trigger events on th
96 system, it is ideal (and customary) to have on
97 CPU core, so that each core can trigger events
98 core.
99
100 You will notice that the clock event device co
101 idea about translating counters to nanoseconds
102 arithmetic, and you find the same family of he
103 assigning these values. The clock event driver
104 attribute however: the system will not try to
105 horizon of the clock event.
106
107
108 sched_clock()
109 -------------
110
111 In addition to the clock sources and clock eve
112 function in the kernel called sched_clock(). T
113 number of nanoseconds since the system was sta
114 may not provide an implementation of sched_clo
115 implementation is not provided, the system jif
116 sched_clock().
117
118 As the name suggests, sched_clock() is used fo
119 determining the absolute timeslice for a certa
120 for example. It is also used for printk timest
121 include time information in printk for things
122
123 Compared to clock sources, sched_clock() has t
124 much more often, especially by the scheduler.
125 between accuracy compared to the clock source,
126 for speed in sched_clock(). It however require
127 characteristics as the clock source, i.e. it s
128
129 The sched_clock() function may wrap only on un
130 i.e. after 64 bits. Since this is a nanosecond
131 after circa 585 years. (For most practical sys
132
133 If an architecture does not provide its own im
134 it will fall back to using jiffies, making its
135 jiffy frequency for the architecture. This wil
136 and will likely show up in system benchmarks.
137
138 The clock driving sched_clock() may stop or re
139 suspend/sleep. This does not matter to the fun
140 events on the system. However it may result in
141 printk().
142
143 The sched_clock() function should be callable
144 NMI-safe and return a sane value in any contex
145
146 Some architectures may have a limited set of t
147 counter to derive a 64-bit nanosecond value, s
148 architecture, special helper functions have be
149 sched_clock() nanosecond base from a 16- or 32
150 same counter that is also used as clock source
151
152 On SMP systems, it is crucial for performance
153 independently on each CPU without any synchron
154 Some hardware (such as the x86 TSC) will cause
155 drift between the CPUs on the system. The kern
156 enabling the CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
157 that makes sched_clock() different from the or
158
159
160 Delay timers (some architectures only)
161 --------------------------------------
162
163 On systems with variable CPU frequency, the va
164 will sometimes behave strangely. Basically the
165 loop to delay a certain number of jiffy fracti
166 jiffy) value, calibrated on boot.
167
168 Let's hope that your system is running on maxi
169 is calibrated: as an effect when the frequency
170 full frequency, any delay() will be twice as l
171 hurt, as you're commonly requesting that amoun
172 basically the semantics are quite unpredictabl
173
174 Enter timer-based delays. Using these, a timer
175 a hard-coded loop for providing the desired de
176
177 This is done by declaring a struct delay_timer
178 function pointers and rate settings for this d
179
180 This is available on some architectures like O
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.