Real Time Clock (RTC) Drivers for Linux

When Linux developers talk about a “Real Time Clock”, they usually meansomething that tracks wall clock time and is battery backed so that itworks even with system power off. Such clocks will normally not trackthe local time zone or daylight savings time – unless they dual bootwith MS-Windows – but will instead be set to Coordinated Universal Time(UTC, formerly “Greenwich Mean Time”).

The newest non-PC hardware tends to just count seconds, like the time(2)system call reports, but RTCs also very commonly represent time usingthe Gregorian calendar and 24 hour time, as reported by gmtime(3).

Linux has two largely-compatible userspace RTC API families you mayneed to know about:

  • /dev/rtc … is the RTC provided by PC compatible systems,so it’s not very portable to non-x86 systems.
  • /dev/rtc0, /dev/rtc1 … are part of a framework that’ssupported by a wide variety of RTC chips on all systems.

Programmers need to understand that the PC/AT functionality is notalways available, and some systems can do much more. That is, theRTCs use the same API to make requests in both RTC frameworks (usingdifferent filenames of course), but the hardware may not offer thesame functionality. For example, not every RTC is hooked up to anIRQ, so they can’t all issue alarms; and where standard PC RTCs canonly issue an alarm up to 24 hours in the future, other hardware maybe able to schedule one any time in the upcoming century.

Old PC/AT-Compatible driver: /dev/rtc

All PCs (even Alpha machines) have a Real Time Clock built into them.Usually they are built into the chipset of the computer, but some mayactually have a Motorola MC146818 (or clone) on the board. This is theclock that keeps the date and time while your computer is turned off.

ACPI has standardized that MC146818 functionality, and extended it ina few ways (enabling longer alarm periods, and wake-from-hibernate).That functionality is NOT exposed in the old driver.

However it can also be used to generate signals from a slow 2Hz to arelatively fast 8192Hz, in increments of powers of two. These signalsare reported by interrupt number 8. (Oh! Sothat is what IRQ 8 isfor…) It can also function as a 24hr alarm, raising IRQ 8 when thealarm goes off. The alarm can also be programmed to only check anysubset of the three programmable values, meaning that it could be set toring on the 30th second of the 30th minute of every hour, for example.The clock can also be set to generate an interrupt upon every clockupdate, thus generating a 1Hz signal.

The interrupts are reported via /dev/rtc (major 10, minor 135, read onlycharacter device) in the form of an unsigned long. The low byte containsthe type of interrupt (update-done, alarm-rang, or periodic) that wasraised, and the remaining bytes contain the number of interrupts sincethe last read. Status information is reported through the pseudo-file/proc/driver/rtc if the /proc filesystem was enabled. The driver hasbuilt in locking so that only one process is allowed to have the /dev/rtcinterface open at a time.

A user process can monitor these interrupts by doing a read(2) or aselect(2) on /dev/rtc – either will block/stop the user process untilthe next interrupt is received. This is useful for things likereasonably high frequency data acquisition where one doesn’t want toburn up 100% CPU by polling gettimeofday etc. etc.

At high frequencies, or under high loads, the user process should checkthe number of interrupts received since the last read to determine ifthere has been any interrupt “pileup” so to speak. Just for reference, atypical 486-33 running a tight read loop on /dev/rtc will start to sufferoccasional interrupt pileup (i.e. > 1 IRQ event since last read) forfrequencies above 1024Hz. So you really should check the high bytesof the value you read, especially at frequencies above that of thenormal timer interrupt, which is 100Hz.

Programming and/or enabling interrupt frequencies greater than 64Hz isonly allowed by root. This is perhaps a bit conservative, but we don’t wantan evil user generating lots of IRQs on a slow 386sx-16, where it might havea negative impact on performance. This 64Hz limit can be changed by writinga different value to /proc/sys/dev/rtc/max-user-freq. Note that theinterrupt handler is only a few lines of code to minimize any possibilityof this effect.

Also, if the kernel time is synchronized with an external source, thekernel will write the time back to the CMOS clock every 11 minutes. Inthe process of doing this, the kernel briefly turns off RTC periodicinterrupts, so be aware of this if you are doing serious work. If youdon’t synchronize the kernel time with an external source (via ntp orwhatever) then the kernel will keep its hands off the RTC, allowing youexclusive access to the device for your applications.

The alarm and/or interrupt frequency are programmed into the RTC viavarious ioctl(2) calls as listed in ./include/linux/rtc.hRather than write 50 pages describing the ioctl() and so on, it isperhaps more useful to include a small test program that demonstrateshow to use them, and demonstrates the features of the driver. This isprobably a lot more useful to people interested in writing applicationsthat will be using this driver. See the code at the end of this document.

(The original /dev/rtc driver was written by Paul Gortmaker.)

New portable “RTC Class” drivers: /dev/rtcN

Because Linux supports many non-ACPI and non-PC platforms, some of whichhave more than one RTC style clock, it needed a more portable solutionthan expecting a single battery-backed MC146818 clone on every system.Accordingly, a new “RTC Class” framework has been defined. It offersthree different userspace interfaces:

  • /dev/rtcN … much the same as the older /dev/rtc interface
  • /sys/class/rtc/rtcN … sysfs attributes support readonlyaccess to some RTC attributes.
  • /proc/driver/rtc … the system clock RTC may expose itselfusing a procfs interface. If there is no RTC for the system clock,rtc0 is used by default. More information is (currently) shownhere than through sysfs.

The RTC Class framework supports a wide variety of RTCs, ranging from thoseintegrated into embeddable system-on-chip (SOC) processors to discrete chipsusing I2C, SPI, or some other bus to communicate with the host CPU. There’seven support for PC-style RTCs … including the features exposed on newer PCsthrough ACPI.

The new framework also removes the “one RTC per system” restriction. Forexample, maybe the low-power battery-backed RTC is a discrete I2C chip, buta high functionality RTC is integrated into the SOC. That system might readthe system clock from the discrete RTC, but use the integrated one for allother tasks, because of its greater functionality.

Check out tools/testing/selftests/rtc/rtctest.c for an example usage of theioctl interface.