blob: ce31f65e12e736130adba6116d1a0a51ff29fff1 [file] [log] [blame]
Thomas Gleixnerdf784882006-01-09 20:52:33 -08001
2hrtimers - subsystem for high-resolution kernel timers
3----------------------------------------------------
4
5This patch introduces a new subsystem for high-resolution kernel timers.
6
7One might ask the question: we already have a timer subsystem
8(kernel/timers.c), why do we need two timer subsystems? After a lot of
9back and forth trying to integrate high-resolution and high-precision
10features into the existing timer framework, and after testing various
11such high-resolution timer implementations in practice, we came to the
12conclusion that the timer wheel code is fundamentally not suitable for
Matt LaPlantefff92892006-10-03 22:47:42 +020013such an approach. We initially didn't believe this ('there must be a way
Thomas Gleixnerdf784882006-01-09 20:52:33 -080014to solve this'), and spent a considerable effort trying to integrate
15things into the timer wheel, but we failed. In hindsight, there are
16several reasons why such integration is hard/impossible:
17
18- the forced handling of low-resolution and high-resolution timers in
19 the same way leads to a lot of compromises, macro magic and #ifdef
20 mess. The timers.c code is very "tightly coded" around jiffies and
21 32-bitness assumptions, and has been honed and micro-optimized for a
22 relatively narrow use case (jiffies in a relatively narrow HZ range)
23 for many years - and thus even small extensions to it easily break
24 the wheel concept, leading to even worse compromises. The timer wheel
25 code is very good and tight code, there's zero problems with it in its
26 current usage - but it is simply not suitable to be extended for
27 high-res timers.
28
29- the unpredictable [O(N)] overhead of cascading leads to delays which
Matt LaPlante992caac2006-10-03 22:52:05 +020030 necessitate a more complex handling of high resolution timers, which
Thomas Gleixnerdf784882006-01-09 20:52:33 -080031 in turn decreases robustness. Such a design still led to rather large
32 timing inaccuracies. Cascading is a fundamental property of the timer
33 wheel concept, it cannot be 'designed out' without unevitably
34 degrading other portions of the timers.c code in an unacceptable way.
35
36- the implementation of the current posix-timer subsystem on top of
37 the timer wheel has already introduced a quite complex handling of
38 the required readjusting of absolute CLOCK_REALTIME timers at
39 settimeofday or NTP time - further underlying our experience by
40 example: that the timer wheel data structure is too rigid for high-res
41 timers.
42
43- the timer wheel code is most optimal for use cases which can be
44 identified as "timeouts". Such timeouts are usually set up to cover
45 error conditions in various I/O paths, such as networking and block
46 I/O. The vast majority of those timers never expire and are rarely
47 recascaded because the expected correct event arrives in time so they
48 can be removed from the timer wheel before any further processing of
49 them becomes necessary. Thus the users of these timeouts can accept
50 the granularity and precision tradeoffs of the timer wheel, and
51 largely expect the timer subsystem to have near-zero overhead.
52 Accurate timing for them is not a core purpose - in fact most of the
53 timeout values used are ad-hoc. For them it is at most a necessary
54 evil to guarantee the processing of actual timeout completions
55 (because most of the timeouts are deleted before completion), which
56 should thus be as cheap and unintrusive as possible.
57
58The primary users of precision timers are user-space applications that
59utilize nanosleep, posix-timers and itimer interfaces. Also, in-kernel
60users like drivers and subsystems which require precise timed events
Matt LaPlante53cb4722006-10-03 22:55:17 +020061(e.g. multimedia) can benefit from the availability of a separate
Thomas Gleixnerdf784882006-01-09 20:52:33 -080062high-resolution timer subsystem as well.
63
64While this subsystem does not offer high-resolution clock sources just
65yet, the hrtimer subsystem can be easily extended with high-resolution
66clock capabilities, and patches for that exist and are maturing quickly.
67The increasing demand for realtime and multimedia applications along
68with other potential users for precise timers gives another reason to
69separate the "timeout" and "precise timer" subsystems.
70
Matt LaPlante53cb4722006-10-03 22:55:17 +020071Another potential benefit is that such a separation allows even more
Thomas Gleixnerdf784882006-01-09 20:52:33 -080072special-purpose optimization of the existing timer wheel for the low
73resolution and low precision use cases - once the precision-sensitive
74APIs are separated from the timer wheel and are migrated over to
75hrtimers. E.g. we could decrease the frequency of the timeout subsystem
76from 250 Hz to 100 HZ (or even smaller).
77
78hrtimer subsystem implementation details
79----------------------------------------
80
81the basic design considerations were:
82
83- simplicity
84
85- data structure not bound to jiffies or any other granularity. All the
86 kernel logic works at 64-bit nanoseconds resolution - no compromises.
87
88- simplification of existing, timing related kernel code
89
90another basic requirement was the immediate enqueueing and ordering of
91timers at activation time. After looking at several possible solutions
92such as radix trees and hashes, we chose the red black tree as the basic
93data structure. Rbtrees are available as a library in the kernel and are
94used in various performance-critical areas of e.g. memory management and
95file systems. The rbtree is solely used for time sorted ordering, while
96a separate list is used to give the expiry code fast access to the
97queued timers, without having to walk the rbtree.
98
Matt LaPlante53cb4722006-10-03 22:55:17 +020099(This separate list is also useful for later when we'll introduce
100high-resolution clocks, where we need separate pending and expired
Thomas Gleixnerdf784882006-01-09 20:52:33 -0800101queues while keeping the time-order intact.)
102
103Time-ordered enqueueing is not purely for the purposes of
104high-resolution clocks though, it also simplifies the handling of
105absolute timers based on a low-resolution CLOCK_REALTIME. The existing
106implementation needed to keep an extra list of all armed absolute
107CLOCK_REALTIME timers along with complex locking. In case of
108settimeofday and NTP, all the timers (!) had to be dequeued, the
109time-changing code had to fix them up one by one, and all of them had to
110be enqueued again. The time-ordered enqueueing and the storage of the
111expiry time in absolute time units removes all this complex and poorly
112scaling code from the posix-timer implementation - the clock can simply
113be set without having to touch the rbtree. This also makes the handling
114of posix-timers simpler in general.
115
116The locking and per-CPU behavior of hrtimers was mostly taken from the
117existing timer wheel code, as it is mature and well suited. Sharing code
118was not really a win, due to the different data structures. Also, the
119hrtimer functions now have clearer behavior and clearer names - such as
120hrtimer_try_to_cancel() and hrtimer_cancel() [which are roughly
121equivalent to del_timer() and del_timer_sync()] - so there's no direct
1221:1 mapping between them on the algorithmical level, and thus no real
123potential for code sharing either.
124
125Basic data types: every time value, absolute or relative, is in a
126special nanosecond-resolution type: ktime_t. The kernel-internal
127representation of ktime_t values and operations is implemented via
128macros and inline functions, and can be switched between a "hybrid
129union" type and a plain "scalar" 64bit nanoseconds representation (at
130compile time). The hybrid union type optimizes time conversions on 32bit
131CPUs. This build-time-selectable ktime_t storage format was implemented
132to avoid the performance impact of 64-bit multiplications and divisions
133on 32bit CPUs. Such operations are frequently necessary to convert
134between the storage formats provided by kernel and userspace interfaces
135and the internal time format. (See include/linux/ktime.h for further
136details.)
137
138hrtimers - rounding of timer values
139-----------------------------------
140
141the hrtimer code will round timer events to lower-resolution clocks
142because it has to. Otherwise it will do no artificial rounding at all.
143
144one question is, what resolution value should be returned to the user by
145the clock_getres() interface. This will return whatever real resolution
146a given clock has - be it low-res, high-res, or artificially-low-res.
147
148hrtimers - testing and verification
149----------------------------------
150
151We used the high-resolution clock subsystem ontop of hrtimers to verify
152the hrtimer implementation details in praxis, and we also ran the posix
153timer tests in order to ensure specification compliance. We also ran
154tests on low-resolution clocks.
155
156The hrtimer patch converts the following kernel functionality to use
157hrtimers:
158
159 - nanosleep
160 - itimers
161 - posix-timers
162
163The conversion of nanosleep and posix-timers enabled the unification of
164nanosleep and clock_nanosleep.
165
166The code was successfully compiled for the following platforms:
167
168 i386, x86_64, ARM, PPC, PPC64, IA64
169
170The code was run-tested on the following platforms:
171
172 i386(UP/SMP), x86_64(UP/SMP), ARM, PPC
173
174hrtimers were also integrated into the -rt tree, along with a
175hrtimers-based high-resolution clock implementation, so the hrtimers
176code got a healthy amount of testing and use in practice.
177
178 Thomas Gleixner, Ingo Molnar