RFC 2783

Network Working Group                                          J. Mogul
Request for Comments: 2783                                   Compaq WRL
Category: Informational                                        D. Mills
                                                 University of Delaware
                                                          J. Brittenson
                                                               J. Stone
                                                               U. Windl
                                                Universitaet Regensburg
                                                             March 2000

   Pulse-Per-Second API for UNIX-like Operating Systems, Version 1.0

Status of this Memo

   This memo provides information for the Internet community.  It does
   not specify an Internet standard of any kind.  Distribution of this
   memo is unlimited.

Copyright Notice

   Copyright (C) The Internet Society (2000).  All Rights Reserved.


   RFC 1589 describes a UNIX kernel implementation model for high-
   precision time-keeping.  This model is meant for use in conjunction
   with the Network Time Protocol (NTP, RFC 1305), or similar time
   synchronization protocols.  One aspect of this model is an accurate
   interface to the high-accuracy, one pulse-per-second (PPS) output
   typically available from precise time sources (such as a GPS or GOES
   receiver).  RFC 1589 did not define an API for managing the PPS
   facility, leaving implementors without a portable means for using PPS
   sources.  This document specifies such an API.

Mogul, et al.                Informational                      [Page 1]

RFC 2783                  Pulse-Per-Second API                March 2000

Table of Contents

   1 Introduction...................................................  2
   2 Data types for representing timestamps.........................  4
   2.1 Resolution...................................................  4
   2.2 Time scale...................................................  5
   3 API............................................................  5
   3.1 PPS abstraction..............................................  6
   3.2 New data structures..........................................  7
   3.3 Mode bit definitions......................................... 10
   3.4 New functions................................................ 12
   3.4.1 New functions: obtaining PPS sources....................... 13
   3.4.2 New functions: setting PPS parameters...................... 14
   3.4.3 New functions: access to PPS timestamps.................... 16
   3.4.4 New functions: disciplining the kernel timebase............ 18
   3.5 Compliance rules............................................. 20
   3.5.1 Functions.................................................. 20
   3.5.2 Mode bits.................................................. 20
   3.6 Examples..................................................... 21
   4 Security Considerations........................................ 24
   5 Acknowledgements............................................... 24
   6 References..................................................... 25
   7 Authors' Addresses............................................. 26
   A. Extensions and related APIs................................... 27
   A.1 Extension: Parameters for the "echo" mechanism............... 27
   A.2 Extension: Obtaining information about external clocks....... 27
   A.3 Extension: Finding a PPS source.............................. 28
   B. Example implementation: PPSDISC Line discipline............... 29
   B.1 Example...................................................... 29
   C. Available implementations..................................... 30
   Full Copyright Statement......................................... 31

1 Introduction

   RFC 1589 [4] describes a model and programming interface for generic
   operating system software that manages the system clock and timer
   functions. The model provides improved accuracy and stability for
   most workstations and servers using the Network Time Protocol (NTP)
   [3] or similar time synchronization protocol.  The model supports the
   use of external timing sources, such as the precision pulse-per-
   second (PPS) signals typically available from precise time sources
   (such as a GPS or GOES receiver).

   However, RFC 1589 did not define an application programming interface
   (API) for the PPS facility.  This document specifies such an
   interface, for use with UNIX (or UNIX-like) operating systems.  Such
   systems often conform to the "Single UNIX Specification" [5],
   sometimes known as POSIX.

Mogul, et al.                Informational                      [Page 2]

RFC 2783                  Pulse-Per-Second API                March 2000

   One convenient means to provide a PPS signal to a computer system is
   to connect that signal to a modem-control pin on a serial-line
   interface to the computer.  The Data Carrier Detect (DCD) pin is
   frequently used for this purpose.  Typically, the time-code output of
   the time source is transmitted to the computer over the same serial
   line.  The computer detects a signal transition on the DCD pin,
   usually by receiving an interrupt, and records a timestamp as soon as

   Although existing practice has focussed on the use of serial lines
   and DCD transitions, PPS signals might also be delivered by other
   kinds of devices.  The API specified in this document does not
   require the use of a serial line, although it may be somewhat biased
   in that direction.

   The typical use of this facility is for the operating system to
   record ("capture") a high-resolution timestamp as soon as possible
   after it detects a PPS signal transition (usually indicated by an
   interrupt).  This timestamp can then be made available, with less
   stringent delay constraints, to time-related software.  The software
   can compare the captured timestamp to the received time-code to
   accurately discover the offset between the system clock and the
   precise time source.

   The operating system may also deliver the PPS event to a kernel
   procedure, called the "in-kernel PPS consumer."  One example would be
   the "hardpps()" procedure, described in RFC 1589, which is used to
   discipline the kernel's internal timebase.

   The API specified in this document allows for one or more signal
   sources attached to a computer system to provide PPS inputs, at the
   option of user-level software.  User-level software may obtain
   signal-transition timestamps for any of these PPS sources.  User-
   level software may optionally specify at most one of these PPS
   sources to be used to discipline the system's internal timebase.

   Although the primary purpose of this API is for capturing true
   pulse-per-second events, the API may also be used for accurately
   timestamping events of other periods, or even aperiodic events, when
   these can be expressed as signal transitions.

   This document does not define internal details of how the API must be
   implemented, and does not specify constraints on the accuracy,
   resolution, or latency of the PPS feature.  However, the utility of
   this feature is inversely proportional to the delay (and variance of
   delay), and implementors are encouraged to take this seriously.

Mogul, et al.                Informational                      [Page 3]

RFC 2783                  Pulse-Per-Second API                March 2000

   In principle, the rate of events to be captured, or the frequency of
   the signals, can range from once per day (or less often) to several
   thousand per second.  However, since in most implementations the
   timestamping function will be implemented as a processor interrupt at
   a relatively high priority, it is prudent to limit the rate of such
   events.  This may be done either by mechanisms in the hardware that
   generates the signals, or by the operating system.

2 Data types for representing timestamps

   Computer systems use various representations of time.  Because this
   API is concerned with the provision of high-accuracy, high-resolution
   time information, the choice of representation is significant.  (Here
   we consider only binary representations, not human-format

   The two interesting questions are:

      1. what is the resolution of the representation?

      2. what time scale is represented?

   These questions often lead to contentious arguments.  Since this API
   is intended for use with NTP and POSIX-compliant systems, however, we
   can limit the choices to representations compatible with existing NTP
   and POSIX practice, even if that practice is considered "wrong" in
   some quarters.

2.1 Resolution

   In the NTP protocol, "timestamps are represented as a 64-bit unsigned
   fixed-point number, in seconds relative to 0h on 1 January 1900. The
   integer part is in the first 32 bits and the fraction part in the
   last 32 bits [...] The precision of this representation is about 200
   picoseconds" [3].

   However, most computer systems cannot measure time to this resolution
   (this represents a clock rate of 5 GHz).  The POSIX gettimeofday()
   function returns a "struct timeval" value, with a resolution of 1
   microsecond.  The POSIX clock_gettime() function returns a "struct
   timespec" value, with a resolution of 1 nanosecond.

   This API uses an extensible representation, but defaults to the
   "struct timespec" representation.

Mogul, et al.                Informational                      [Page 4]

RFC 2783                  Pulse-Per-Second API                March 2000

2.2 Time scale

   Several different time scales have been proposed for use in computer
   systems.  UTC and TAI are the two obvious candidates.

   Some people would prefer the use of TAI, which is identical to UTC
   except that it does not correct for leap seconds.  Their preference
   for TAI stems from the difficulty of computing precise time
   differences when leap seconds are involved, especially when using
   times in the future (for which the exact number of leap seconds is,
   in general, unknowable).

   However, POSIX and NTP both use UTC, albeit with different base
   dates.  Given that support for TAI would, in general, require other
   changes to the POSIX specification, this API uses the POSIX base date
   of 00:00 January 1, 1970 UTC, and conforms to the POSIX use of the
   UTC time scale.


   A PPS facility can be used in two different ways:

      1. An application can obtain a timestamp, using the system's
         internal timebase, for the most recent PPS event.

      2. The kernel may directly utilize PPS events to discipline its
         internal timebase, thereby providing highly accurate time to
         all applications.

   This API supports both uses, individually or in combination.  The
   timestamping feature may be used on any number of PPS sources
   simultaneously; the timebase-disciplining feature may be used with at
   most one PPS source.

   Although the proper implementation of this API requires support from
   the kernel of a UNIX system, this document defines the API in terms
   of a set of library routines.  This gives the implementor some
   freedom to divide the effort between kernel code and library code
   (different divisions might be appropriate on microkernels and
   monolithic kernels, for example).

Mogul, et al.                Informational                      [Page 5]

RFC 2783                  Pulse-Per-Second API                March 2000

3.1 PPS abstraction

   A PPS signal consists of a series of pulses, each with an "asserted"
   (logical true) phase, and a "clear" (logical false) phase.  The two
   phases may be of different lengths.  The API may capture an "assert
   timestamp" at the moment of the transition into the asserted phase,
   and a "clear timestamp" at the moment of the transition into the
   clear phase.

   The specific assignment of the logical values "true" and "false" with
   specific voltages of a PPS signal, if applicable, is outside the
   scope of this specification.  However, these assignments SHOULD be
   consistent with applicable standards.  Implementors of PPS sources
   SHOULD document these assignments.

      Reminder to implementors of DCD-based PPS support:  TTL and RS-
      232C (V.24/V.28) interfaces both define the "true" state as the
      one having the highest positive voltage. TTL defines a nominal
      absence of voltage as the "false" state, but RS-232C (V.24/V.28)
      defines the "false" state by the presence of a negative voltage.

   The API supports the direct provision of PPS events (and timestamps)
   to an in-kernel PPS consumer.  This could be the function called
   "hardpps()", as described in RFC 1589 [4], but the API does not
   require the kernel implementation to use that function name
   internally.  The current version of the API supports at most one in-
   kernel PPS consumer, and does not provide a way to explicitly name
   it.  The implementation SHOULD impose access controls on the use of
   this feature.

   The API optionally supports an "echo" feature, in which events on the
   incoming PPS signal may be reflected through software, after the
   capture of the corresponding timestamp, to an output signal pin.
   This feature may be used to discover an upper bound on the actual
   delay between the edges of the PPS signal and the capture of the
   timestamps; such information may be useful in precise calibration of
   the system.

   The designation of an output pin for the echo signal, and sense and
   shape of the output transition, is outside the scope of this
   specification, but SHOULD be documented for each implementation.  The
   output pin MAY also undergo transitions at other times besides those
   caused by PPS input events.

      Note: this allows an implementation of the echo feature to
      generate an output pulse per input pulse, or an output edge per
      input pulse, or an output pulse per input edge. It also allows the
      same signal pin to be used for several purposes simultaneously.

Mogul, et al.                Informational                      [Page 6]

RFC 2783                  Pulse-Per-Second API                March 2000

   Also, the API optionally provides an application with the ability to
   specify an offset value to be applied to captured timestamps.  This
   can be used to correct for cable and/or radio-wave propagation
   delays, or to compensate for systematic jitter in the external
   signal.  The implementation SHOULD impose access controls on the use
   of this feature.

3.2 New data structures

   The data structure declarations and symbol definitions for this API
   will appear in the header file <sys/timepps.h>.  The header file MUST
   define all constants described in this specification, even if they
   are not supported by the implementation.

   The API includes several implementation-specific types:

      typedef ... pps_handle_t;       /* represents a PPS source */

      typedef unsigned ... pps_seq_t; /* sequence number */

   The "pps_handle_t" type is an opaque scalar type used to represent a
   PPS source within the API.

   The "pps_seq_t" type is an unsigned integer data type of at least 32

   The precise declaration of the pps_handle_t and pps_seq_t types is

   The API imports the standard POSIX definition for this data type:

      struct timespec {
              time_t  tv_sec;         /* seconds */
              long    tv_nsec;        /* nanoseconds */

   The API defines this structure as an internal (not "on the wire")
   representation of the NTP "64-bit unsigned fixed-point" timestamp
   format [3]:

      typedef struct ntp_fp {
              unsigned int    integral;
              unsigned int    fractional;
      } ntp_fp_t;

   The two fields in this structure may be declared as any unsigned
   integral type, each of at least 32 bits.

Mogul, et al.                Informational                      [Page 7]

RFC 2783                  Pulse-Per-Second API                March 2000

   The API defines this new union as an extensible type for representing

      typedef union pps_timeu {
              struct timespec tspec;
              ntp_fp_t        ntpfp;
              unsigned long   longpad[3];
      } pps_timeu_t;

   Future revisions of this specification may add more fields to this

      Note: adding a field to this union that is larger than
      3*sizeof(long) will break binary compatibility.

   The API defines these new data structures:

      typedef struct {
          pps_seq_t   assert_sequence;        /* assert event seq # */
          pps_seq_t   clear_sequence;         /* clear event seq # */
          pps_timeu_t assert_tu;
          pps_timeu_t clear_tu;
          int         current_mode;           /* current mode bits */
      } pps_info_t;

      #define assert_timestamp        assert_tu.tspec
      #define clear_timestamp         clear_tu.tspec

      #define assert_timestamp_ntpfp  assert_tu.ntpfp
      #define clear_timestamp_ntpfp   clear_tu.ntpfp

      typedef struct {
          int         api_version;            /* API version # */
          int         mode;                   /* mode bits */
          pps_timeu_t assert_off_tu;
          pps_timeu_t clear_off_tu;
      } pps_params_t;

      #define assert_offset   assert_off_tu.tspec
      #define clear_offset    clear_off_tu.tspec

      #define assert_offset_ntpfp     assert_off_tu.ntpfp
      #define clear_offset_ntpfp      clear_off_tu.ntpfp

   The "pps_info_t" type is returned on an inquiry to PPS source.  It
   contains the timestamps for the most recent assert event, and the
   most recent clear event.  The order in which these events were
   actually received is defined by the timetamps, not by any other

Mogul, et al.                Informational                      [Page 8]

RFC 2783                  Pulse-Per-Second API                March 2000

   aspect of the specification.  Each timestamp field represents the
   value of the operating system's internal timebase when the
   timestamped event occurred, or as close as possible to that time
   (with the optional addition of a specified offset).  The current_mode
   field contains the value of the mode bits (see section 3.3) at the
   time of the most recent transition was captured for this PPS source.
   An application can use current_mode to discover the format of the
   timestamps returned.

   The assert_sequence number increases once per captured assert
   timestamp.  Its initial value is undefined.  If incremented past the
   largest value for the type, the next value is zero.  The
   clear_sequence number increases once per captured clear timestamp.
   Its initial value is undefined, and may be different from the initial
   value of assert_sequence.  If incremented past the largest value for
   the type, the next value is zero.  Due to possible signal loss or
   excessive signal noise, the assert-sequence number and the clear-
   sequence number might not always increase in step with each other.

      Note that these sequence numbers are most useful in applications
      where events other than PPS transitions are to be captured, which
      might be involved in a precision stopwatch application, for
      example. In such cases, the sequence numbers may be used to detect
      overruns, where the application has missed one or more events.
      They may also be used to detect an excessive event rate, or to
      detect that an event has failed to occur between two calls to the
      time_pps_fetch() function (defined later).

      In order to obtain an uninterrupted series of sequence numbers
      (and hence of event timestamps), it may be necessary to sample the
      pps_info_t values at a rate somewhat faster than the underlying
      event rate.  For example, an application interested in both assert
      and clear timestamps may need to sample at least twice per second.
      Proper use of the sequence numbers allows an application to
      discover if it has missed any event timestamps due to an
      insufficient sampling rate.

   The pps_params_t data type is used to discover and modify parameters
   of a PPS source.  The data type includes a mode field, described in
   section 3.3.  It also includes an api_version field, a read-only
   value giving the version of the API.  Currently, the only defined
   value is:

      #define PPS_API_VERS_1  1

   This field is present to enable binary compatibility with future
   versions of the API.

Mogul, et al.                Informational                      [Page 9]

RFC 2783                  Pulse-Per-Second API                March 2000

      Note: the term "read-only" in this specification means that an
      application cannot modify the relevant data item; only the
      implementation can modify the value.  The implementation MUST
      ignore attempts by the application to modify a read-only field.

   As an OPTIONAL feature of the API, the implementation MAY support
   adding offsets to the timestamps that are captured.  (Values of type
   "struct timespec" can represent negative offsets.)  The assert_offset
   field of a pps_params_t value specifies a value to be added to
   generate a captured assert_timestamp.  The clear_offset of a
   pps_params_t value field specifies a value to be added to generate a
   captured clear_timestamp.  Since the offsets, if any, apply to all
   users of a given PPS source, the implementation SHOULD impose access
   controls on the use of this feature; for example, allowing only the
   super-user to set the offset values.  The default value for both
   offsets is zero.

3.3 Mode bit definitions

   A set of mode bits is associated with each PPS source.

   The bits in the mode field of the pps_params_t type are:

      /* Device/implementation parameters */
      #define PPS_CAPTUREASSERT       0x01
      #define PPS_CAPTURECLEAR        0x02
      #define PPS_CAPTUREBOTH         0x03

      #define PPS_OFFSETASSERT        0x10
      #define PPS_OFFSETCLEAR         0x20

      #define PPS_CANWAIT             0x100
      #define PPS_CANPOLL             0x200

      /* Kernel actions */
      #define PPS_ECHOASSERT          0x40
      #define PPS_ECHOCLEAR           0x80

      /* Timestamp formats */
      #define PPS_TSFMT_TSPEC         0x1000
      #define PPS_TSFMT_NTPFP         0x2000

   These mode bits are divided into three categories:

      1. Device/implementation parameters:  These are parameters either
         of the device or of the implementation.  If the implementation
         allows these to be changed, then these bits are read/write for
         users with sufficient privilege (such as the super-user), and

Mogul, et al.                Informational                     [Page 10]

RFC 2783                  Pulse-Per-Second API                March 2000

         read-only for other users.  If the implementation does not
         allow these bits to be changed, they are read-only.

      2. Kernel actions:  These bits specify certain kernel actions to
         be taken on arrival of a signal.  If the implementation
         supports one of these actions, then the corresponding bit is
         read/write for users with sufficient privilege (such as the
         super-user), and read-only for other users.  If the
         implementation does not support the action, the corresponding
         bit is always zero.

      3. Timestamp formats:  These bits indicate the set of timestamp
         formats available for the device.  They are always read-only.

   In more detail, the meanings of the Device/implementation parameter
   mode bits are:

                   If this bit is set, the assert timestamp for the
                   associated PPS source will be captured.

                   If this bit is set, the clear timestamp for the
                   associated PPS source will be captured.

   PPS_CAPTUREBOTH Defined as the union of PPS_CAPTUREASSERT and
                   PPS_CAPTURECLEAR, for convenience.

                   If set, the assert_offset value is added to the
                   current value of the operating system's internal
                   timebase in order to generate the captured

   PPS_OFFSETCLEAR If set, the clear_offset value is added to the
                   current value of the operating system's internal
                   timebase in order to generate the captured

   PPS_CANWAIT     If set, the application may request that the
                   time_pps_fetch() function (see section 3.4.3) should
                   block until the next timestamp arrives.  Note: this
                   mode bit is read-only.

   PPS_CANPOLL     This bit is reserved for future use.  An application
                   SHOULD NOT depend on any functionality implied either
                   by its presence or by its absence.

Mogul, et al.                Informational                     [Page 11]

RFC 2783                  Pulse-Per-Second API                March 2000

   If neither PPS_CAPTUREASSERT nor PPS_CAPTURECLEAR is set, no valid
   timestamp will be available via the API.

   The meanings of the Kernel action mode bits are:

   PPS_ECHOASSERT   If set, after the capture of an assert timestamp,
                   the implementation generates a signal transition as
                   rapidly as possible on an output signal pin.  This
                   MUST NOT affect the delay between the PPS source's
                   transition to the asserted phase and the capture of
                   the assert timestamp.

   PPS_ECHOCLEAR    If set, after the capture of a clear timestamp, the
                   implementation generates a signal transition as
                   rapidly as possible on an output signal pin.  This
                   MUST NOT affect the delay between the PPS source's
                   transition to the clear phase and the capture of the
                   clear timestamp.

   The timestamp formats are:

   PPS_TSFMT_TSPEC Timestamps and offsets are represented as values of
                   type "struct timespec".  All implementations MUST
                   support this format, and this format is the default
                   unless an application specifies otherwise.

   PPS_TSFMT_NTPFP Timestamps and offsets are represented as values of
                   type "ntp_fp_t", which corresponds to the NTP
                   "64-bit unsigned fixed-point" timestamp format [3].
                   Support for this format is OPTIONAL.

   Other timestamp format bits may be defined as fields are added to the
   "pps_timeu_t" union.

   The operating system may implement all of these mode bits, or just a
   subset of them.  If an attempt is made to set an unsupported mode
   bit, the API will return an error.  If an attempt is made to modify a
   read-only mode bit, the API will return an error.

3.4 New functions

   In the description of functions that follows, we use the following
   function parameters:

   filedes         A file descriptor (type: int), for a serial line or
                   other source of PPS events.

Mogul, et al.                Informational                     [Page 12]

RFC 2783                  Pulse-Per-Second API                March 2000

   ppshandle       A variable of type "pps_handle_t", as defined in
                   section 3.2.

   ppsinfobuf      A record of type "pps_info_t", as defined in
                   section 3.2.

   ppsparams       A record of type "pps_params_t", as defined in
                   section 3.2.

   tsformat        An integer with exactly one of the timestamp format
                   bits set.

3.4.1 New functions: obtaining PPS sources

   The API includes functions to create and destroy PPS source


      int time_pps_create(int filedes, pps_handle_t *handle);
      int time_pps_destroy(pps_handle_t handle);


   All of the other functions in the PPS API operate on PPS handles
   (type: pps_handle_t).  The time_pps_create() is used to convert an
   already-open UNIX file descriptor, for an appropriate special file,
   into a PPS handle.

   The definition of what special files are appropriate for use with the
   PPS API is outside the scope of this specification, and may vary
   based on both operating system implementation, and local system
   configuration.  One typical case is a serial line, whose DCD pin is
   connected to a source of PPS events.

   The mode in which the UNIX file descriptor was originally opened
   affects what operations are allowed on the PPS handle.  The
   time_pps_setparams() and time_pps_kcbind() functions (see sections
   3.4.2 and 3.4.4) SHOULD be prohibited by the implementation if the
   descriptor is open only for reading (O_RDONLY).

      Note: operations on a descriptor opened with an inappropriate mode
      might fail with EBADF.

   The time_pps_destroy() function makes the PPS handle unusable, and
   frees any storage that might have been allocated for it.  It does not
   close the associated file descriptor, nor does it change any of the
   parameter settings for the PPS source.

Mogul, et al.                Informational                     [Page 13]

RFC 2783                  Pulse-Per-Second API                March 2000

      Note: If this API is adapted to an operating system that does not
      follow UNIX conventions for representing an accessible PPS source
      as an integer file descriptor, the time_pps_create() function may
      take different parameters from those shown here.


   On successful completion, the time_pps_create() function returns 0.
   Otherwise, a value of -1 is returned and errno is set to indicate the

   If called with a valid handle parameter, the time_pps_destroy()
   function returns 0.  Otherwise, it returns -1.


   If the time_pps_create() function fails, errno may be set to one of
   the following values:

   [EBADF]         The filedes parameter is not a valid file descriptor.

   [EOPNOTSUPP]    The use of the PPS API is not supported for the file

   [EPERM]         The process's effective user ID does not have the
                   required privileges to use the PPS API.

3.4.2 New functions: setting PPS parameters

   The API includes several functions use to set or obtain the
   parameters of a PPS source.


      int time_pps_setparams(pps_handle_t handle,
                              const pps_params_t *ppsparams);
      int time_pps_getparams(pps_handle_t handle,
                              pps_params_t *ppsparams);
      int time_pps_getcap(pps_handle_t handle, int *mode);


   A suitably privileged application may use time_pps_setparams() to set
   the parameters (mode bits and timestamp offsets) for a PPS source.
   The pps_params_t type is defined in section 3.2; mode bits are
   defined in section 3.3.  An application may use time_pps_getparams()
   to discover the current settings of the PPS parameters.  An
   application that needs to change only a subset of the existing

Mogul, et al.                Informational                     [Page 14]

RFC 2783                  Pulse-Per-Second API                March 2000

   parameters must first call time_pps_getparams() to obtain the current
   parameter values, then set the new values using time_pps_setparams().

      Note: a call to time_pps_setparams() replaces the current values
      of all mode bits with those specified via the ppsparams argument,
      except those bits whose state cannot be changed.  Bits might be
      read-only due to access controls, or because they are fixed by the

   The timestamp format of the assert_offset and clear_offset fields is
   defined by the mode field.  That is, on a call to
   time_pps_setparams(), the kernel interprets the supplied offset
   values using the timestamp format given in the mode field of the
   ppsparams argument.  If the requested timestamp format is not
   supported, the time_pps_setparams() function has no effect and
   returns an error value.  On a call to time_pps_getparams(), the
   kernel provides the timestamp format of the offsets by setting one of
   the timestamp format bits in the mode field.

      Note: an application that uses time_pps_getparams() to read the
      current offset values cannot specify which format is used.  The
      implementation SHOULD return the offsets using the same timestamp
      format as was used when the offsets were set.

   An application wishing to discover which mode bits it may set, with
   its current effective user ID, may call time_pps_getcap().  This
   function returns the set of mode bits that may be set by the
   application, without generating an EINVAL or EPERM error, for the
   specified PPS source.  It does not return the current values for the
   mode bits.  A call to time_pps_getcap() returns the mode bits
   corresponding to all supported timestamp formats.

   The time_pps_getcap() function MAY ignore the mode in which the
   associated UNIX file descriptor was opened, so the application might
   still receive an EBADF error on a call to time_pps_setparams(), even
   if time_pps_getcap() says that the chosen mode bits are allowed.

   The mode bits returned by time_pps_getcap() for distinct PPS handles
   may differ, reflecting the specific capabilities of the underlying
   hardware connection to the PPS source, or of the source itself.


   On successful completion, the time_pps_setparams(),
   time_pps_getparams(), and time_pps_getcap() functions return 0.
   Otherwise, a value of -1 is returned and errno is set to indicate the

Mogul, et al.                Informational                     [Page 15]

RFC 2783                  Pulse-Per-Second API                March 2000


   If the time_pps_setparams(), time_pps_getparams(), or
   time_pps_getcap() function fails, errno may be set to one of the
   following values:

   [EBADF]         The handle parameter is not associated with a valid
                   file descriptor, or the descriptor is not open for

   [EFAULT]        A parameter points to an invalid address.

   [EOPNOTSUPP]    The use of the PPS API is not supported for the
                   associated file descriptor.

   [EINVAL]        The operating system does not support all of the
                   requested mode bits.

   [EPERM]         The process's effective user ID does not have the
                   required privileges to use the PPS API, or to set the
                   given mode bits.

3.4.3 New functions: access to PPS timestamps

   The API includes one function that gives applications access to PPS
   timestamps.  As an implementation option, the application may request
   the API to block until the next timestamp is captured.  (The API does
   not directly support the use of the select() or poll() system calls
   to wait for PPS events.)


      int time_pps_fetch(pps_handle_t handle,
                              const int tsformat,
                              pps_info_t *ppsinfobuf,
                              const struct timespec *timeout);


   An application may use time_pps_fetch() to obtain the most recent
   timestamps captured for the PPS source specified by the handle
   parameter.  The tsformat parameter specifies the desired timestamp
   format; if the requested timestamp format is not supported, the call
   fails and returns an error value.  The application MUST specify
   exactly one timestamp format.

Mogul, et al.                Informational                     [Page 16]

RFC 2783                  Pulse-Per-Second API                March 2000

   This function blocks until either a timestamp is captured from the
   PPS source, or until the specified timeout duration has expired.  If
   the timeout parameter is a NULL pointer, the function simply blocks
   until a timestamp is captured.  If the timeout parameter specifies a
   delay of zero, the function returns immediately.

   Support for blocking behavior is an implementation option.  If the
   PPS_CANWAIT mode bit is clear, and the timeout parameter is either
   NULL or points to a non-zero value, the function returns an
   EOPNOTSUPP error.  An application can discover whether the feature is
   implemented by using time_pps_getcap() to see if the PPS_CANWAIT mode
   bit is set.

   The result is stored in the ppsinfobuf parameter, whose fields are
   defined in section 3.2.  If the function returns as the result of a
   timeout or error, the contents of the ppsinfobuf are undefined.

   If this function is invoked before the system has captured a
   timestamp for the signal source, the ppsinfobuf returned will have
   its timestamp fields set to the time format's base date (e.g., for
   PPS_TSFMT_TSPEC, both the tv_sec and tv_nsec fields will be zero).


   On successful completion, the time_pps_fetch() function returns 0.
   Otherwise, a value of -1 is returned and errno is set to indicate the


   If the time_pps_fetch() function fails, errno may be set to one of
   the following values:

   [EBADF]         The handle parameter is not associated with a valid
                   file descriptor.

   [EFAULT]        A parameter points to an invalid address.

   [EINTR]         A signal was delivered before the time limit
                   specified by the timeout parameter expired and before
                   a timestamp has been captured.

   [EINVAL]        The requested timestamp format is not supported.

   [EOPNOTSUPP]    The use of the PPS API is not supported for the
                   associated file descriptor.

   [ETIMEDOUT]     The timeout duration has expired.

Mogul, et al.                Informational                     [Page 17]

RFC 2783                  Pulse-Per-Second API                March 2000

3.4.4 New functions: disciplining the kernel timebase

   The API includes one OPTIONAL function to specify if and how a PPS
   source is provided to a kernel consumer of PPS events, such as the
   code used to discipline the operating system's internal timebase.


      int time_pps_kcbind(pps_handle_t handle,
                              const int kernel_consumer,
                              const int edge,
                              const int tsformat);

   An application with appropriate privileges may use time_pps_kcbind()
   to bind a kernel consumer to the PPS source specified by the handle.

   The kernel consumer is identified by the kernel_consumer parameter.
   In the current version of the API, the possible values for this
   parameter are:

      #define PPS_KC_HARDPPS          0
      #define PPS_KC_HARDPPS_PLL      1
      #define PPS_KC_HARDPPS_FLL      2

   with these meanings:

   PPS_KC_HARDPPS  The kernel's hardpps() function (or equivalent).

                   A variant of hardpps() constrained to use a
                   phase-locked loop.

                   A variant of hardpps() constrained to use a
                   frequency-locked loop.

   Implementation of any or all of these values is OPTIONAL.

   The edge parameter indicates which edge of the PPS signal causes a
   timestamp to be delivered to the kernel consumer.  It may have the
   depending on particular characteristics of the PPS source.  It may
   also be zero, which removes any binding between the PPS source and
   the kernel consumer.

Mogul, et al.                Informational                     [Page 18]

RFC 2783                  Pulse-Per-Second API                March 2000

   The tsformat parameter specifies the format for the timestamps
   delivered to the kernel consumer.  If this value is zero, the
   implementation MAY choose the appropriate format, or return EINVAL.
   The implementation MAY ignore a non-zero value for this parameter.

   The binding created by this call persists until it is changed by a
   subsequent call specifying the same kernel_consumer.  In particular,
   a subsequent call to time_pps_destroy() for the specified handle does
   not affect the binding.

   The binding is independent of any prior or subsequent changes to the
   PPS_CAPTUREASSERT and PPS_CAPTURECLEAR mode bits for the device.
   However, if either the edge or the tsformat parameter values are
   inconsistent with the capabilities of the PPS source, an error is
   returned.  The implementation MAY also return an error if the
   tsformat value is unsupported for time_pps_kcbind(), even if it is
   supported for other uses of the API.

   The operating system may enforce two restrictions on the bindings
   created by time_pps_kcbind():

      1. the kernel MAY return an error if an attempt is made to bind a
         kernel consumer to more than one PPS source a time.

      2. the kernel MAY restrict the ability to set bindings to
         processes with sufficient privileges to modify the system's
         internal timebase.  (On UNIX systems, such modification is
         normally done using settimeofday() and/or adjtime(), and is
         restricted to users with superuser privilege.)

      Warning: If this feature is configured for a PPS source that does
      not have an accurate 1-pulse-per-second signal, or is otherwise
      inappropriately configured, use of this feature may result in
      seriously incorrect timekeeping for the entire system.  For best
      results, the 1-PPS signal should have much better frequency
      stability than the system's internal clock source (usually a
      crystal-controlled oscillator), and should have jitter (variation
      in interarrival time) much less than the system's clock-tick

   See RFC 1589 [4] for more information about how the system's timebase
   may be disciplined using a PPS signal.


   On successful completion, the time_pps_kcbind() function returns 0.
   Otherwise, a value of -1 is returned and errno is set to indicate the

Mogul, et al.                Informational                     [Page 19]

RFC 2783                  Pulse-Per-Second API                March 2000


   If the time_pps_kcbind() function fails, errno may be set to one of
   the following values:

   [EBADF]         The handle parameter is not associated with a valid
                   file descriptor, or the descriptor is not open for

   [EFAULT]        A parameter points to an invalid address.

   [EINVAL]        The requested timestamp format is not supported.

   [EOPNOTSUPP]    The use of the PPS API is not supported for the
                   associated file descriptor, or this OPTIONAL
                   function is not supported.

   [EPERM]         The process's effective user ID does not have the
                   required privileges to set the binding.

3.5 Compliance rules

   The key words "MUST", "MUST NOT", "REQUIRED","SHOULD", SHOULD NOT",
   "MAY", and "OPTIONAL" in this document are to be interpreted as
   described in RFC 2119 [1].

   Some features of this specification are OPTIONAL, but others are

3.5.1 Functions

   An implementation MUST provide these functions:

      - time_pps_create()
      - time_pps_destroy()
      - time_pps_setparams()
      - time_pps_getparams()
      - time_pps_getcap()
      - time_pps_fetch()

   An implementation MUST provide this function, but it may be
   implemented as a function that always return an EOPNOTSUPP error,
   possibly on a per-source basis:

      - time_pps_kcbind()

Mogul, et al.                Informational                     [Page 20]

RFC 2783                  Pulse-Per-Second API                March 2000

3.5.2 Mode bits

   An implementation MUST support at least one of these mode bits for
   each PPS source:


   and MAY support both of them.  If an implementation supports both of
   these bits for a PPS source, it SHOULD allow them to be set

   An implementation MUST support this timestamp format:


   An implementation MAY support these mode bits:


   An implementation MAY support this timestamp format:


3.6 Examples

   A very simple use of this API might be:

      int fd;
      pps_handle_t handle;
      pps_params_t params;
      pps_info_t infobuf;
      struct timespec timeout;

      /* Open a file descriptor and enable PPS on rising edges */
      fd = open(PPSfilename, O_RDWR, 0);
      time_pps_create(fd, &handle);
      time_pps_getparams(handle, ¶ms);
      if ((params.mode & PPS_CAPTUREASSERT) == 0) {
          fprintf(stderr, "%s cannot currently CAPTUREASSERT\n",

      /* create a zero-valued timeout */

Mogul, et al.                Informational                     [Page 21]

RFC 2783                  Pulse-Per-Second API                March 2000

      timeout.tv_sec = 0;
      timeout.tv_nsec = 0;

      /* loop, printing the most recent timestamp every second or so */
      while (1) {
          time_pps_fetch(handle, PPS_TSFMT_TSPEC, &infobuf, &timeout);
          printf("Assert timestamp: %d.%09d, sequence: %ld\n",

   Note that this example omits most of the error-checking that would be
   expected in a reliable program.

   Also note that, on a system that supports PPS_CANWAIT, the function
   of these lines:

         time_pps_fetch(handle, PPS_TSFMT_TSPEC, &infobuf, &timeout);

   might be more reliably accomplished using:

         timeout.tv_sec = 100;
         timeout.tv_nsec = 0;
         time_pps_fetch(handle, PPS_TSFMT_TSPEC, &infobuf, &timeout);

   The (arbitrary) timeout value is used to protect against the
   possibility that another application might disable PPS timestamps, or
   that the hardware generating the timestamps might fail.

   A slightly more elaborate use of this API might be:

      int fd;
      pps_handle_t handle;
      pps_params_t params;
      pps_info_t infobuf;
      int avail_mode;
      struct timespec timeout;

      /* Open a file descriptor */
      fd = open(PPSfilename, O_RDWR, 0);
      time_pps_create(fd, &handle);

       * Find out what features are supported

Mogul, et al.                Informational                     [Page 22]

RFC 2783                  Pulse-Per-Second API                March 2000

      time_pps_getcap(handle, &avail_mode);
      if ((avail_mode & PPS_CAPTUREASSERT) == 0) {
          fprintf(stderr, "%s cannot CAPTUREASSERT\n", PPSfilename);
      if ((avail_mode & PPS_OFFSETASSERT) == 0) {
          fprintf(stderr, "%s cannot OFFSETASSERT\n", PPSfilename);

       * Capture assert timestamps, and
       *   compensate for a 675 nsec propagation delay

      time_pps_getparams(handle, ¶ms);
      params.assert_offset.tv_sec = 0;
      params.assert_offset.tv_nsec = 675;
      time_pps_setparams(handle, ¶ms);

      /* create a zero-valued timeout */
      timeout.tv_sec = 0;
      timeout.tv_nsec = 0;

      /* loop, printing the most recent timestamp every second or so */
      while (1) {
          if (avail_mode & PPS_CANWAIT) {
              time_pps_fetch(handle, PPS_TSFMT_TSPEC, &infobuf, NULL);
                              /* waits for the next event */
          } else {
              time_pps_fetch(handle, PPS_TSFMT_TSPEC, &infobuf,

          printf("Assert timestamp: %d.%09d, sequence: %ld\n",

   Again, most of the necessary error-checking has been omitted from
   this example.

Mogul, et al.                Informational                     [Page 23]

RFC 2783                  Pulse-Per-Second API                March 2000

4 Security Considerations

   This API gives applications three capabilities:

      - Causing the system to capture timestamps on certain events.

      - Obtaining timestamps for certain events.

      - Affecting the system's internal timebase.

   The first capability should not affect security directly, but might
   cause a slight increase in interrupt latency and interrupt-handling

   The second capability might be useful in implementing certain kinds
   of covert communication channels.

   In most cases, neither of these first two issues is a significant
   security threat, because the traditional UNIX file protection
   facility may be used to to limit access to the relevant special
   files.  Provision of the PPS API adds minimal additional risk.

   The final capability is reserved to highly privileged users.  In UNIX
   systems, this means those with superuser privilege.  Such users can
   evade protections based on file permissions; however, such users can
   in general cause unbounded havoc, and can set the internal timebase
   (and its rate of change), so this API creates no new vulnerabilities.

5 Acknowledgements

   The API in this document draws some of its inspiration from the LBL
   "ppsclock" distribution [2], originally implemented in 1993 by Steve
   McCanne, Craig Leres, and Van Jacobson.  We also thank Poul-Henning
   Kamp, Craig Leres, Judah Levine, and Harlan Stenn for helpful
   comments they contributed during the drafting of this document.

Mogul, et al.                Informational                     [Page 24]

RFC 2783                  Pulse-Per-Second API                March 2000

6 References

   1.  Bradner, S., "Key words for use in RFCs to Indicate Requirement
       Levels", BCP 14, RFC 2119, March 1997.

   2.  Steve McCanne, Craig Leres, and Van Jacobson.  PPSCLOCK.

   3.  Mills, D., "Network Time Protocol (Version 3):  Specification,
       Implementation and Analysis", RFC 1305, March 1992.

   4.  Mills, D., "A Kernel Model for Precision Timekeeping", RFC 1589,
       March, 1994.

   5.  The Open Group.  The Single UNIX Specification, Version 2 - 6 Vol
       Set for UNIX 98.  Document number T912, The Open Group, February,

Mogul, et al.                Informational                     [Page 25]

RFC 2783                  Pulse-Per-Second API                March 2000

7 Authors' Addresses

   Jeffrey C. Mogul
   Western Research Laboratory
   Compaq Computer Corporation
   250 University Avenue
   Palo Alto, California, 94305, U.S.A.

   Phone: 1 650 617 3304 (email preferred)
   EMail: mogul@wrl.dec.com

   David L. Mills
   Electrical and Computer Engineering Department
   University of Delaware
   Newark, DE 19716

   Phone: (302) 831-8247
   EMail: mills@udel.edu

   Jan Brittenson
   Sun Microsystems, Inc.
   901 San Antonio Rd  M/S MPK17-202
   Palo Alto, CA 94303
   Email: Jan.Brittenson@Eng.Sun.COM

   Jonathan Stone
   Stanford Distributed Systems Group
   Stanford, CA 94305

   Phone: (650) 723-2513
   EMail: jonathan@dsg.stanford.edu

   Ulrich Windl
   Universitaet Regensburg, Klinikum

   EMail: ulrich.windl@rz.uni-regensburg.de

Mogul, et al.                Informational                     [Page 26]

RFC 2783                  Pulse-Per-Second API                March 2000

A. Extensions and related APIs

   The API specified in the main body of this document could be more
   useful with the provision of several extensions or companion APIs.

   At present, the interfaces listed in this appendix are not part of
   the formal specification in this document.

A.1 Extension: Parameters for the "echo" mechanism

   The "echo" mechanism described in the body of this specification
   leaves most of the details to the implementor, especially the
   designation of one or more output pins.

   It might be useful to extend this API to provide either or both of
   these features:

      - A means by which the application can discover which output
        pin is echoing the input pin.

      - A means by which the application can select which output
        pin is echoing the input pin.

A.2 Extension: Obtaining information about external clocks

   The PPS API may be useful with a wide variety of reference clocks,
   connected via several different interface technologies (including
   serial lines, parallel interfaces, and bus-level interfaces).  These
   reference clocks can have many features and parameters, some of which
   might not even have been invented yet.

   We believe that it would be useful to have a mechanism by which an
   application can discover arbitrary features and parameters of a
   reference clock.  These might include:

      - Clock manufacturer, model number, and revision level
      - Whether the clock is synchronized to an absolute standard
      - For synchronized clocks,
           * The specific standard
           * The accuracy of the standard
           * The path used (direct connection, shortwave, longwave,
             satellite, etc.)
           * The distance (offset) and variability of this path

Mogul, et al.                Informational                     [Page 27]

RFC 2783                  Pulse-Per-Second API                March 2000

      - For PPS sources,
           * The pulse rate
           * The pulse shape
           * Which edge of the pulse corresponds to the epoch

      - The time representation format

   This information might best be provided by an API analogous to the
   standard "curses" API, with a database analogous to the standard
   "terminfo" database.  That is, a "clockinfo" database would contain a
   set of (attribute, value) pairs for each type of clock, and the API
   would provide a means to query this database.

   Additional mechanisms would allow an application to discover the
   clock or clocks connected to the local system, and to discover the
   clockinfo type of a specific clock device.

A.3 Extension: Finding a PPS source

   Although the clockinfo database described in section A.2, together
   with the discover mechanisms described there, would allow an
   application to discover the PPS source (or sources) connected to a
   system, it might be more complex than necessary.

   A simpler approach would be to support a single function that
   provides the identity of one or more PPS sources.

   For example, the function might be declared as

      int time_pps_findsource(int index,
                              char *path, int pathlen,
                              char *idstring, int idlen);

   The index argument implicitly sets up an ordering on the PPS sources
   attached to the system.  An application would use this function to
   inquire about the Nth source.  The function would return -1 if no
   such source exists; otherwise, it would return 0, and would place the
   pathname of the associated special file in the path argument.  It
   would also place an identification string in the idstring argument.
   The identification string could include the clock make, model,
   version, etc., which could then be used by the application to control
   its behavior.

   This function might simply read the Nth line from a simple database,
   containing lines such as:

      /dev/tty00      "TrueTime 468-DC"
      /dev/pps1       "Homebrew rubidium frequency standard"

Mogul, et al.                Informational                     [Page 28]

RFC 2783                  Pulse-Per-Second API                March 2000

   allowing the system administrator to describe the configuration of
   PPS sources.

B. Example implementation: PPSDISC Line discipline

   One possible implementation of the PPS API might be to define a new
   "line discipline" and then map the API onto a set of ioctl()
   commands.  Here we sketch such an implementation; note that this is
   not part of the specification of the API, and applications should not
   expect this low-level interface to be available.

   In this approach, the set of line disciplines is augmented with one
   new line discipline, PPSDISC.  This discipline will act exactly the
   same as the TTYDISC discipline, except for its handling of modem DCD

   Once the TIOCSETD ioctl() has been used to select this line
   discipline, PPS-related operations on the serial line may be invoked
   using new ioctl() commands.  For example (values used only for

   #define PPSFETCH      _IOR('t', 75, pps_info_t)
   #define PPSSETPARAM   _IOW('t', 76, pps_params_t)
   #define PPSGETPARAM   _IOR('t', 77, pps_params_t)
   #define PPSGETCAP     _IOR('t', 78, int)

B.1 Example

   A typical use might be:

      int ldisc = PPSDISC;
      pps_params_t params;
      pps_info_t infobuf;

      ioctl(fd, TIOCSETD, &ldisc);    /* set discipline */

       * Check the capabilities of this PPS source to see
       * if it supports what we need.
      ioctl(fd, PPSGETCAP, ¶ms);
      if ((params.mode & PPS_CAPTUREASSERT) == 0) {
          fprintf(stderr, "PPS source is not suitable\n");

       * Set this line to timestamp on a rising-edge interrupt

Mogul, et al.                Informational                     [Page 29]

RFC 2783                  Pulse-Per-Second API                March 2000

      ioctl(fd, PPSGETPARAMS, ¶ms);
      params.mode |= PPS_CAPTUREASSERT;
      ioctl(fd, PPSSETPARAMS, ¶ms);

      sleep(2);       /* allow time for the PPS pulse to happen */

      /* obtain most recent timestamp and sequence # for this line */
      ioctl(fd, PPSFETCH, &infobuf);

   Again, this example imprudently omits any error-checking.

C. Available implementations

   Several available implementations of this API are listed at
   <http://www.ntp.org/ppsapi/PPSImpList.html>.  Note that not all of
   these implementations correspond to the current version of the

Mogul, et al.                Informational                     [Page 30]

RFC 2783                  Pulse-Per-Second API                March 2000

Full Copyright Statement

   Copyright (C) The Internet Society (2000).  All Rights Reserved.

   This document and translations of it may be copied and furnished to
   others, and derivative works that comment on or otherwise explain it
   or assist in its implementation may be prepared, copied, published
   and distributed, in whole or in part, without restriction of any
   kind, provided that the above copyright notice and this paragraph are
   included on all such copies and derivative works.  However, this
   document itself may not be modified in any way, such as by removing
   the copyright notice or references to the Internet Society or other
   Internet organizations, except as needed for the purpose of
   developing Internet standards in which case the procedures for
   copyrights defined in the Internet Standards process must be
   followed, or as required to translate it into languages other than

   The limited permissions granted above are perpetual and will not be
   revoked by the Internet Society or its successors or assigns.

   This document and the information contained herein is provided on an


   Funding for the RFC Editor function is currently provided by the
   Internet Society.

Mogul, et al.                Informational                     [Page 31]