1 ===================================================== 2 High resolution timers and dynamic ticks design notes 3 ===================================================== 4 5 Further information can be found in the paper of the OLS 2006 talk "hrtimers 6 and beyond". The paper is part of the OLS 2006 Proceedings Volume 1, which can 7 be found on the OLS website: 8 https://www.kernel.org/doc/ols/2006/ols2006v1-pages-333-346.pdf 9 10 The slides to this talk are available from: 11 http://www.cs.columbia.edu/~nahum/w6998/papers/ols2006-hrtimers-slides.pdf 12 13 The slides contain five figures (pages 2, 15, 18, 20, 22), which illustrate the 14 changes in the time(r) related Linux subsystems. Figure #1 (p. 2) shows the 15 design of the Linux time(r) system before hrtimers and other building blocks 16 got merged into mainline. 17 18 Note: the paper and the slides are talking about "clock event source", while we 19 switched to the name "clock event devices" in meantime. 20 21 The design contains the following basic building blocks: 22 23 - hrtimer base infrastructure 24 - timeofday and clock source management 25 - clock event management 26 - high resolution timer functionality 27 - dynamic ticks 28 29 30 hrtimer base infrastructure 31 --------------------------- 32 33 The hrtimer base infrastructure was merged into the 2.6.16 kernel. Details of 34 the base implementation are covered in Documentation/timers/hrtimers.rst. See 35 also figure #2 (OLS slides p. 15) 36 37 The main differences to the timer wheel, which holds the armed timer_list type 38 timers are: 39 40 - time ordered enqueueing into a rb-tree 41 - independent of ticks (the processing is based on nanoseconds) 42 43 44 timeofday and clock source management 45 ------------------------------------- 46 47 John Stultz's Generic Time Of Day (GTOD) framework moves a large portion of 48 code out of the architecture-specific areas into a generic management 49 framework, as illustrated in figure #3 (OLS slides p. 18). The architecture 50 specific portion is reduced to the low level hardware details of the clock 51 sources, which are registered in the framework and selected on a quality based 52 decision. The low level code provides hardware setup and readout routines and 53 initializes data structures, which are used by the generic time keeping code to 54 convert the clock ticks to nanosecond based time values. All other time keeping 55 related functionality is moved into the generic code. The GTOD base patch got 56 merged into the 2.6.18 kernel. 57 58 Further information about the Generic Time Of Day framework is available in the 59 OLS 2005 Proceedings Volume 1: 60 61 http://www.linuxsymposium.org/2005/linuxsymposium_procv1.pdf 62 63 The paper "We Are Not Getting Any Younger: A New Approach to Time and 64 Timers" was written by J. Stultz, D.V. Hart, & N. Aravamudan. 65 66 Figure #3 (OLS slides p.18) illustrates the transformation. 67 68 69 clock event management 70 ---------------------- 71 72 While clock sources provide read access to the monotonically increasing time 73 value, clock event devices are used to schedule the next event 74 interrupt(s). The next event is currently defined to be periodic, with its 75 period defined at compile time. The setup and selection of the event device 76 for various event driven functionalities is hardwired into the architecture 77 dependent code. This results in duplicated code across all architectures and 78 makes it extremely difficult to change the configuration of the system to use 79 event interrupt devices other than those already built into the 80 architecture. Another implication of the current design is that it is necessary 81 to touch all the architecture-specific implementations in order to provide new 82 functionality like high resolution timers or dynamic ticks. 83 84 The clock events subsystem tries to address this problem by providing a generic 85 solution to manage clock event devices and their usage for the various clock 86 event driven kernel functionalities. The goal of the clock event subsystem is 87 to minimize the clock event related architecture dependent code to the pure 88 hardware related handling and to allow easy addition and utilization of new 89 clock event devices. It also minimizes the duplicated code across the 90 architectures as it provides generic functionality down to the interrupt 91 service handler, which is almost inherently hardware dependent. 92 93 Clock event devices are registered either by the architecture dependent boot 94 code or at module insertion time. Each clock event device fills a data 95 structure with clock-specific property parameters and callback functions. The 96 clock event management decides, by using the specified property parameters, the 97 set of system functions a clock event device will be used to support. This 98 includes the distinction of per-CPU and per-system global event devices. 99 100 System-level global event devices are used for the Linux periodic tick. Per-CPU 101 event devices are used to provide local CPU functionality such as process 102 accounting, profiling, and high resolution timers. 103 104 The management layer assigns one or more of the following functions to a clock 105 event device: 106 107 - system global periodic tick (jiffies update) 108 - cpu local update_process_times 109 - cpu local profiling 110 - cpu local next event interrupt (non periodic mode) 111 112 The clock event device delegates the selection of those timer interrupt related 113 functions completely to the management layer. The clock management layer stores 114 a function pointer in the device description structure, which has to be called 115 from the hardware level handler. This removes a lot of duplicated code from the 116 architecture specific timer interrupt handlers and hands the control over the 117 clock event devices and the assignment of timer interrupt related functionality 118 to the core code. 119 120 The clock event layer API is rather small. Aside from the clock event device 121 registration interface it provides functions to schedule the next event 122 interrupt, clock event device notification service and support for suspend and 123 resume. 124 125 The framework adds about 700 lines of code which results in a 2KB increase of 126 the kernel binary size. The conversion of i386 removes about 100 lines of 127 code. The binary size decrease is in the range of 400 byte. We believe that the 128 increase of flexibility and the avoidance of duplicated code across 129 architectures justifies the slight increase of the binary size. 130 131 The conversion of an architecture has no functional impact, but allows to 132 utilize the high resolution and dynamic tick functionalities without any change 133 to the clock event device and timer interrupt code. After the conversion the 134 enabling of high resolution timers and dynamic ticks is simply provided by 135 adding the kernel/time/Kconfig file to the architecture specific Kconfig and 136 adding the dynamic tick specific calls to the idle routine (a total of 3 lines 137 added to the idle function and the Kconfig file) 138 139 Figure #4 (OLS slides p.20) illustrates the transformation. 140 141 142 high resolution timer functionality 143 ----------------------------------- 144 145 During system boot it is not possible to use the high resolution timer 146 functionality, while making it possible would be difficult and would serve no 147 useful function. The initialization of the clock event device framework, the 148 clock source framework (GTOD) and hrtimers itself has to be done and 149 appropriate clock sources and clock event devices have to be registered before 150 the high resolution functionality can work. Up to the point where hrtimers are 151 initialized, the system works in the usual low resolution periodic mode. The 152 clock source and the clock event device layers provide notification functions 153 which inform hrtimers about availability of new hardware. hrtimers validates 154 the usability of the registered clock sources and clock event devices before 155 switching to high resolution mode. This ensures also that a kernel which is 156 configured for high resolution timers can run on a system which lacks the 157 necessary hardware support. 158 159 The high resolution timer code does not support SMP machines which have only 160 global clock event devices. The support of such hardware would involve IPI 161 calls when an interrupt happens. The overhead would be much larger than the 162 benefit. This is the reason why we currently disable high resolution and 163 dynamic ticks on i386 SMP systems which stop the local APIC in C3 power 164 state. A workaround is available as an idea, but the problem has not been 165 tackled yet. 166 167 The time ordered insertion of timers provides all the infrastructure to decide 168 whether the event device has to be reprogrammed when a timer is added. The 169 decision is made per timer base and synchronized across per-cpu timer bases in 170 a support function. The design allows the system to utilize separate per-CPU 171 clock event devices for the per-CPU timer bases, but currently only one 172 reprogrammable clock event device per-CPU is utilized. 173 174 When the timer interrupt happens, the next event interrupt handler is called 175 from the clock event distribution code and moves expired timers from the 176 red-black tree to a separate double linked list and invokes the softirq 177 handler. An additional mode field in the hrtimer structure allows the system to 178 execute callback functions directly from the next event interrupt handler. This 179 is restricted to code which can safely be executed in the hard interrupt 180 context. This applies, for example, to the common case of a wakeup function as 181 used by nanosleep. The advantage of executing the handler in the interrupt 182 context is the avoidance of up to two context switches - from the interrupted 183 context to the softirq and to the task which is woken up by the expired 184 timer. 185 186 Once a system has switched to high resolution mode, the periodic tick is 187 switched off. This disables the per system global periodic clock event device - 188 e.g. the PIT on i386 SMP systems. 189 190 The periodic tick functionality is provided by an per-cpu hrtimer. The callback 191 function is executed in the next event interrupt context and updates jiffies 192 and calls update_process_times and profiling. The implementation of the hrtimer 193 based periodic tick is designed to be extended with dynamic tick functionality. 194 This allows to use a single clock event device to schedule high resolution 195 timer and periodic events (jiffies tick, profiling, process accounting) on UP 196 systems. This has been proved to work with the PIT on i386 and the Incrementer 197 on PPC. 198 199 The softirq for running the hrtimer queues and executing the callbacks has been 200 separated from the tick bound timer softirq to allow accurate delivery of high 201 resolution timer signals which are used by itimer and POSIX interval 202 timers. The execution of this softirq can still be delayed by other softirqs, 203 but the overall latencies have been significantly improved by this separation. 204 205 Figure #5 (OLS slides p.22) illustrates the transformation. 206 207 208 dynamic ticks 209 ------------- 210 211 Dynamic ticks are the logical consequence of the hrtimer based periodic tick 212 replacement (sched_tick). The functionality of the sched_tick hrtimer is 213 extended by three functions: 214 215 - hrtimer_stop_sched_tick 216 - hrtimer_restart_sched_tick 217 - hrtimer_update_jiffies 218 219 hrtimer_stop_sched_tick() is called when a CPU goes into idle state. The code 220 evaluates the next scheduled timer event (from both hrtimers and the timer 221 wheel) and in case that the next event is further away than the next tick it 222 reprograms the sched_tick to this future event, to allow longer idle sleeps 223 without worthless interruption by the periodic tick. The function is also 224 called when an interrupt happens during the idle period, which does not cause a 225 reschedule. The call is necessary as the interrupt handler might have armed a 226 new timer whose expiry time is before the time which was identified as the 227 nearest event in the previous call to hrtimer_stop_sched_tick. 228 229 hrtimer_restart_sched_tick() is called when the CPU leaves the idle state before 230 it calls schedule(). hrtimer_restart_sched_tick() resumes the periodic tick, 231 which is kept active until the next call to hrtimer_stop_sched_tick(). 232 233 hrtimer_update_jiffies() is called from irq_enter() when an interrupt happens 234 in the idle period to make sure that jiffies are up to date and the interrupt 235 handler has not to deal with an eventually stale jiffy value. 236 237 The dynamic tick feature provides statistical values which are exported to 238 userspace via /proc/stat and can be made available for enhanced power 239 management control. 240 241 The implementation leaves room for further development like full tickless 242 systems, where the time slice is controlled by the scheduler, variable 243 frequency profiling, and a complete removal of jiffies in the future. 244 245 246 Aside the current initial submission of i386 support, the patchset has been 247 extended to x86_64 and ARM already. Initial (work in progress) support is also 248 available for MIPS and PowerPC. 249 250 Thomas, Ingo
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.