181 lines · plain
1===========================================================2Clock sources, Clock events, sched_clock() and delay timers3===========================================================4 5This document tries to briefly explain some basic kernel timekeeping6abstractions. It partly pertains to the drivers usually found in7drivers/clocksource in the kernel tree, but the code may be spread out8across the kernel.9 10If you grep through the kernel source you will find a number of architecture-11specific implementations of clock sources, clockevents and several likewise12architecture-specific overrides of the sched_clock() function and some13delay timers.14 15To provide timekeeping for your platform, the clock source provides16the basic timeline, whereas clock events shoot interrupts on certain points17on this timeline, providing facilities such as high-resolution timers.18sched_clock() is used for scheduling and timestamping, and delay timers19provide an accurate delay source using hardware counters.20 21 22Clock sources23-------------24 25The purpose of the clock source is to provide a timeline for the system that26tells you where you are in time. For example issuing the command 'date' on27a Linux system will eventually read the clock source to determine exactly28what time it is.29 30Typically the clock source is a monotonic, atomic counter which will provide31n bits which count from 0 to (2^n)-1 and then wraps around to 0 and start over.32It will ideally NEVER stop ticking as long as the system is running. It33may stop during system suspend.34 35The clock source shall have as high resolution as possible, and the frequency36shall be as stable and correct as possible as compared to a real-world wall37clock. It should not move unpredictably back and forth in time or miss a few38cycles here and there.39 40It must be immune to the kind of effects that occur in hardware where e.g.41the counter register is read in two phases on the bus lowest 16 bits first42and the higher 16 bits in a second bus cycle with the counter bits43potentially being updated in between leading to the risk of very strange44values from the counter.45 46When the wall-clock accuracy of the clock source isn't satisfactory, there47are various quirks and layers in the timekeeping code for e.g. synchronizing48the user-visible time to RTC clocks in the system or against networked time49servers using NTP, but all they do basically is update an offset against50the clock source, which provides the fundamental timeline for the system.51These measures does not affect the clock source per se, they only adapt the52system to the shortcomings of it.53 54The clock source struct shall provide means to translate the provided counter55into a nanosecond value as an unsigned long long (unsigned 64 bit) number.56Since this operation may be invoked very often, doing this in a strict57mathematical sense is not desirable: instead the number is taken as close as58possible to a nanosecond value using only the arithmetic operations59multiply and shift, so in clocksource_cyc2ns() you find:60 61 ns ~= (clocksource * mult) >> shift62 63You will find a number of helper functions in the clock source code intended64to aid in providing these mult and shift values, such as65clocksource_khz2mult(), clocksource_hz2mult() that help determine the66mult factor from a fixed shift, and clocksource_register_hz() and67clocksource_register_khz() which will help out assigning both shift and mult68factors using the frequency of the clock source as the only input.69 70For real simple clock sources accessed from a single I/O memory location71there is nowadays even clocksource_mmio_init() which will take a memory72location, bit width, a parameter telling whether the counter in the73register counts up or down, and the timer clock rate, and then conjure all74necessary parameters.75 76Since a 32-bit counter at say 100 MHz will wrap around to zero after some 4377seconds, the code handling the clock source will have to compensate for this.78That is the reason why the clock source struct also contains a 'mask'79member telling how many bits of the source are valid. This way the timekeeping80code knows when the counter will wrap around and can insert the necessary81compensation code on both sides of the wrap point so that the system timeline82remains monotonic.83 84 85Clock events86------------87 88Clock events are the conceptual reverse of clock sources: they take a89desired time specification value and calculate the values to poke into90hardware timer registers.91 92Clock events are orthogonal to clock sources. The same hardware93and register range may be used for the clock event, but it is essentially94a different thing. The hardware driving clock events has to be able to95fire interrupts, so as to trigger events on the system timeline. On an SMP96system, it is ideal (and customary) to have one such event driving timer per97CPU core, so that each core can trigger events independently of any other98core.99 100You will notice that the clock event device code is based on the same basic101idea about translating counters to nanoseconds using mult and shift102arithmetic, and you find the same family of helper functions again for103assigning these values. The clock event driver does not need a 'mask'104attribute however: the system will not try to plan events beyond the time105horizon of the clock event.106 107 108sched_clock()109-------------110 111In addition to the clock sources and clock events there is a special weak112function in the kernel called sched_clock(). This function shall return the113number of nanoseconds since the system was started. An architecture may or114may not provide an implementation of sched_clock() on its own. If a local115implementation is not provided, the system jiffy counter will be used as116sched_clock().117 118As the name suggests, sched_clock() is used for scheduling the system,119determining the absolute timeslice for a certain process in the CFS scheduler120for example. It is also used for printk timestamps when you have selected to121include time information in printk for things like bootcharts.122 123Compared to clock sources, sched_clock() has to be very fast: it is called124much more often, especially by the scheduler. If you have to do trade-offs125between accuracy compared to the clock source, you may sacrifice accuracy126for speed in sched_clock(). It however requires some of the same basic127characteristics as the clock source, i.e. it should be monotonic.128 129The sched_clock() function may wrap only on unsigned long long boundaries,130i.e. after 64 bits. Since this is a nanosecond value this will mean it wraps131after circa 585 years. (For most practical systems this means "never".)132 133If an architecture does not provide its own implementation of this function,134it will fall back to using jiffies, making its maximum resolution 1/HZ of the135jiffy frequency for the architecture. This will affect scheduling accuracy136and will likely show up in system benchmarks.137 138The clock driving sched_clock() may stop or reset to zero during system139suspend/sleep. This does not matter to the function it serves of scheduling140events on the system. However it may result in interesting timestamps in141printk().142 143The sched_clock() function should be callable in any context, IRQ- and144NMI-safe and return a sane value in any context.145 146Some architectures may have a limited set of time sources and lack a nice147counter to derive a 64-bit nanosecond value, so for example on the ARM148architecture, special helper functions have been created to provide a149sched_clock() nanosecond base from a 16- or 32-bit counter. Sometimes the150same counter that is also used as clock source is used for this purpose.151 152On SMP systems, it is crucial for performance that sched_clock() can be called153independently on each CPU without any synchronization performance hits.154Some hardware (such as the x86 TSC) will cause the sched_clock() function to155drift between the CPUs on the system. The kernel can work around this by156enabling the CONFIG_HAVE_UNSTABLE_SCHED_CLOCK option. This is another aspect157that makes sched_clock() different from the ordinary clock source.158 159 160Delay timers (some architectures only)161--------------------------------------162 163On systems with variable CPU frequency, the various kernel delay() functions164will sometimes behave strangely. Basically these delays usually use a hard165loop to delay a certain number of jiffy fractions using a "lpj" (loops per166jiffy) value, calibrated on boot.167 168Let's hope that your system is running on maximum frequency when this value169is calibrated: as an effect when the frequency is geared down to half the170full frequency, any delay() will be twice as long. Usually this does not171hurt, as you're commonly requesting that amount of delay *or more*. But172basically the semantics are quite unpredictable on such systems.173 174Enter timer-based delays. Using these, a timer read may be used instead of175a hard-coded loop for providing the desired delay.176 177This is done by declaring a struct delay_timer and assigning the appropriate178function pointers and rate settings for this delay timer.179 180This is available on some architectures like OpenRISC or ARM.181