2 RFC: Common Trace Format Proposal for Linux (v1.6)
4 Mathieu Desnoyers, EfficiOS Inc.
6 The goal of the present document is to propose a trace format that suits the
7 needs of the embedded, telecom, high-performance and kernel communities. It is
8 based on the Common Trace Format Requirements (v1.4) document. It is designed to
9 allow tracing that is natively generated by the Linux kernel and Linux
10 user-space applications written in C/C++.
12 A reference implementation of a library to read and write this trace format is
13 being implemented within the BabelTrace project, a converter between trace
14 formats. The development tree is available at:
16 git tree: git://git.efficios.com/babeltrace.git
17 gitweb: http://git.efficios.com/?p=babeltrace.git
20 1. Preliminary definitions
22 - Event Trace: An ordered sequence of events.
23 - Event Stream: An ordered sequence of events, containing a subset of the
25 - Event Packet: A sequence of physically contiguous events within an event
27 - Event: This is the basic entry in a trace. (aka: a trace record).
28 - An event identifier (ID) relates to the class (a type) of event within
30 e.g. event: irq_entry.
31 - An event (or event record) relates to a specific instance of an event
33 e.g. event: irq_entry, at time X, on CPU Y
34 - Source Architecture: Architecture writing the trace.
35 - Reader Architecture: Architecture reading the trace.
38 2. High-level representation of a trace
40 A trace is divided into multiple event streams. Each event stream contains a
41 subset of the trace event types.
43 The final output of the trace, after its generation and optional transport over
44 the network, is expected to be either on permanent or temporary storage in a
45 virtual file system. Because each event stream is appended to while a trace is
46 being recorded, each is associated with a separate file for output. Therefore,
47 a stored trace can be represented as a directory containing one file per stream.
49 A metadata event stream contains information on trace event types. It describes:
53 - Per-stream event header description.
54 - Per-stream event header selection.
55 - Per-stream event context fields.
57 - Event type to stream mapping.
58 - Event type to name mapping.
59 - Event type to ID mapping.
60 - Event fields description.
65 An event stream is divided in contiguous event packets of variable size. These
66 subdivisions have a variable size. An event packet can contain a certain amount
67 of padding at the end. The rationale for the event stream design choices is
68 explained in Appendix B. Stream Header Rationale.
70 An event stream is divided in contiguous event packets of variable size. These
71 subdivisions have a variable size. An event packet can contain a certain amount
72 of padding at the end. The stream header is repeated at the beginning of each
75 The event stream header will therefore be referred to as the "event packet
76 header" throughout the rest of this document.
83 A basic type is a scalar type, as described in this section.
85 4.1.1 Type inheritance
87 Type specifications can be inherited to allow deriving types from a
88 type class. For example, see the uint32_t named type derived from the "integer"
89 type class below ("Integers" section). Types have a precise binary
90 representation in the trace. A type class has methods to read and write these
91 types, but must be derived into a type to be usable in an event field.
95 We define "byte-packed" types as aligned on the byte size, namely 8-bit.
96 We define "bit-packed" types as following on the next bit, as defined by the
99 All basic types, except bitfields, are either aligned on an architecture-defined
100 specific alignment or byte-packed, depending on the architecture preference.
101 Architectures providing fast unaligned write byte-packed basic types to save
102 space, aligning each type on byte boundaries (8-bit). Architectures with slow
103 unaligned writes align types on specific alignment values. If no specific
104 alignment is declared for a type nor its parents, it is assumed to be bit-packed
105 for bitfields and byte-packed for other types.
107 Metadata attribute representation of a specific alignment:
109 align = value; /* value in bits */
113 By default, the native endianness of the source architecture the trace is used.
114 Byte order can be overridden for a basic type by specifying a "byte_order"
115 attribute. Typical use-case is to specify the network byte order (big endian:
116 "be") to save data captured from the network into the trace without conversion.
117 If not specified, the byte order is native.
119 Metadata representation:
121 byte_order = native OR network OR be OR le; /* network and be are aliases */
125 Type size, in bits, for integers and floats is that returned by "sizeof()" in C
126 multiplied by CHAR_BIT.
127 We require the size of "char" and "unsigned char" types (CHAR_BIT) to be fixed
128 to 8 bits for cross-endianness compatibility.
130 Metadata representation:
132 size = value; (value is in bits)
136 Signed integers are represented in two-complement. Integer alignment, size,
137 signedness and byte ordering are defined in the metadata. Integers aligned on
138 byte size (8-bit) and with length multiple of byte size (8-bit) correspond to
139 the C99 standard integers. In addition, integers with alignment and/or size that
140 are _not_ a multiple of the byte size are permitted; these correspond to the C99
141 standard bitfields, with the added specification that the CTF integer bitfields
142 have a fixed binary representation. A MIT-licensed reference implementation of
143 the CTF portable bitfields is available at:
145 http://git.efficios.com/?p=babeltrace.git;a=blob;f=include/babeltrace/bitfield.h
147 Binary representation of integers:
149 - On little and big endian:
150 - Within a byte, high bits correspond to an integer high bits, and low bits
151 correspond to low bits.
153 - Integer across multiple bytes are placed from the less significant to the
155 - Consecutive integers are placed from lower bits to higher bits (even within
158 - Integer across multiple bytes are placed from the most significant to the
160 - Consecutive integers are placed from higher bits to lower bits (even within
163 This binary representation is derived from the bitfield implementation in GCC
164 for little and big endian. However, contrary to what GCC does, integers can
165 cross units boundaries (no padding is required). Padding can be explicitely
166 added (see 4.1.6 GNU/C bitfields) to follow the GCC layout if needed.
168 Metadata representation:
171 signed = true OR false; /* default false */
172 byte_order = native OR network OR be OR le; /* default native */
173 size = value; /* value in bits, no default */
174 align = value; /* value in bits */
177 Example of type inheritance (creation of a uint32_t named type):
185 Definition of a named 5-bit signed bitfield:
193 4.1.6 GNU/C bitfields
195 The GNU/C bitfields follow closely the integer representation, with a
196 particularity on alignment: if a bitfield cannot fit in the current unit, the
197 unit is padded and the bitfield starts at the following unit. The unit size is
198 defined by the size of the type "unit_type".
200 Metadata representation. Either:
203 unit_type = integer {
209 Or bitfield within structures as specified by the C standard
213 As an example, the following structure declared in C compiled by GCC:
220 is equivalent to the following structure declaration, aligned on the largest
221 element (short). The second bitfield would be aligned on the next unit boundary,
222 because it would not fit in the current unit. The two declarations (C
223 declaration above or CTF declaration with "type gcc_bitfield") are strictly
239 The floating point values byte ordering is defined in the metadata.
241 Floating point values follow the IEEE 754-2008 standard interchange formats.
242 Description of the floating point values include the exponent and mantissa size
243 in bits. Some requirements are imposed on the floating point values:
245 - FLT_RADIX must be 2.
246 - mant_dig is the number of digits represented in the mantissa. It is specified
247 by the ISO C99 standard, section 5.2.4, as FLT_MANT_DIG, DBL_MANT_DIG and
248 LDBL_MANT_DIG as defined by <float.h>.
249 - exp_dig is the number of digits represented in the exponent. Given that
250 mant_dig is one bit more than its actual size in bits (leading 1 is not
251 needed) and also given that the sign bit always takes one bit, exp_dig can be
254 - sizeof(float) * CHAR_BIT - FLT_MANT_DIG
255 - sizeof(double) * CHAR_BIT - DBL_MANT_DIG
256 - sizeof(long double) * CHAR_BIT - LDBL_MANT_DIG
258 Metadata representation:
263 byte_order = native OR network OR be OR le;
266 Example of type inheritance:
268 typedef floating_point {
269 exp_dig = 8; /* sizeof(float) * CHAR_BIT - FLT_MANT_DIG */
270 mant_dig = 24; /* FLT_MANT_DIG */
274 TODO: define NaN, +inf, -inf behavior.
278 Enumerations are a mapping between an integer type and a table of strings. The
279 numerical representation of the enumeration follows the integer type specified
280 by the metadata. The enumeration mapping table is detailed in the enumeration
281 description within the metadata. The mapping table maps inclusive value ranges
282 (or single values) to strings. Instead of being limited to simple
283 "value -> string" mappings, these enumerations map
284 "[ start_value ... end_value ] -> string", which map inclusive ranges of
285 values to strings. An enumeration from the C language can be represented in
286 this format by having the same start_value and end_value for each element, which
287 is in fact a range of size 1. This single-value range is supported without
288 repeating the start and end values with the value = string declaration. If the
289 <integer_type> is omitted, the type chosen by the C compiler to hold the
290 enumeration is used. The <integer_type> specifier can only be omitted for
291 enumerations containing only simple "value -> string" mappings (compatible with
294 enum <integer_type> name {
295 string = start_value1 ... end_value1,
296 "other string" = start_value2 ... end_value2,
297 yet_another_string, /* will be assigned to end_value2 + 1 */
298 "some other string" = value,
302 If the values are omitted, the enumeration starts at 0 and increment of 1 for
313 Overlapping ranges within a single enumeration are implementation defined.
319 Structures are aligned on the largest alignment required by basic types
320 contained within the structure. (This follows the ISO/C standard for structures)
322 Metadata representation of a named structure:
325 field_type field_name;
326 field_type field_name;
333 integer { /* Nameless type */
338 uint64_t second_field_name; /* Named type declared in the metadata */
341 The fields are placed in a sequence next to each other. They each possess a
342 field name, which is a unique identifier within the structure.
344 A nameless structure can be declared as a field type:
352 Arrays are fixed-length. Their length is declared in the type declaration within
353 the metadata. They contain an array of "inner type" elements, which can refer to
354 any type not containing the type of the array being declared (no circular
355 dependency). The length is the number of elements in an array.
357 Metadata representation of a named array, either:
366 typedef elem_type name[length];
372 elem_type = uint32_t;
375 A nameless array can be declared as a field type, e.g.:
384 uint8_t field_name[10];
389 Sequences are dynamically-sized arrays. They start with an integer that specify
390 the length of the sequence, followed by an array of "inner type" elements.
391 The length is the number of elements in the sequence.
393 Metadata representation for a named sequence, either:
396 length_type = type; /* integer class */
402 typedef elem_type name[length_type];
404 A nameless sequence can be declared as a field type, e.g.:
413 long field_name[int];
415 The length type follows the integer types specifications, and the sequence
416 elements follow the "array" specifications.
420 Strings are an array of bytes of variable size and are terminated by a '\0'
421 "NULL" character. Their encoding is described in the metadata. In absence of
422 encoding attribute information, the default encoding is UTF-8.
424 Metadata representation of a named string type:
427 encoding = UTF8 OR ASCII;
430 A nameless string type can be declared as a field type:
432 string field_name; /* Use default UTF8 encoding */
434 5. Event Packet Header
436 The event packet header consists of two part: one is mandatory and have a fixed
437 layout. The second part, the "event packet context", has its layout described in
440 - Aligned on page size. Fixed size. Fields either aligned or packed (depending
441 on the architecture preference).
442 No padding at the end of the event packet header. Native architecture byte
445 Fixed layout (event packet header):
447 - Magic number (CTF magic numbers: 0xC1FC1FC1 and its reverse endianness
448 representation: 0xC11FFCC1) It needs to have a non-symmetric bytewise
449 representation. Used to distinguish between big and little endian traces (this
450 information is determined by knowing the endianness of the architecture
451 reading the trace and comparing the magic number against its value and the
452 reverse, 0xC11FFCC1). This magic number specifies that we use the CTF metadata
453 description language described in this document. Different magic numbers
454 should be used for other metadata description languages.
455 - Trace UUID, used to ensure the event packet match the metadata used.
456 (note: we cannot use a metadata checksum because metadata can be appended to
457 while tracing is active)
458 - Stream ID, used as reference to stream description in metadata.
460 Metadata-defined layout (event packet context):
462 - Event packet content size (in bytes).
463 - Event packet size (in bytes, includes padding).
464 - Event packet content checksum (optional). Checksum excludes the event packet
466 - Per-stream event packet sequence count (to deal with UDP packet loss). The
467 number of significant sequence counter bits should also be present, so
468 wrap-arounds are deal with correctly.
469 - Timestamp at the beginning and timestamp at the end of the event packet.
470 Both timestamps are written in the packet header, but sampled respectively
471 while (or before) writing the first event and while (or after) writing the
472 last event in the packet. The inclusive range between these timestamps should
473 include all event timestamps assigned to events contained within the packet.
474 - Events discarded count
475 - Snapshot of a per-stream free-running counter, counting the number of
476 events discarded that were supposed to be written in the stream prior to
477 the first event in the event packet.
478 * Note: producer-consumer buffer full condition should fill the current
479 event packet with padding so we know exactly where events have been
481 - Lossless compression scheme used for the event packet content. Applied
482 directly to raw data. New types of compression can be added in following
483 versions of the format.
484 0: no compression scheme
488 - Cypher used for the event packet content. Applied after compression.
491 - Checksum scheme used for the event packet content. Applied after encryption.
497 5.1 Event Packet Header Fixed Layout Description
499 struct event_packet_header {
501 uint8_t trace_uuid[16];
505 5.2 Event Packet Context Description
507 Event packet context example. These are declared within the stream declaration
508 in the metadata. All these fields are optional except for "content_size" and
509 "packet_size", which must be present in the context.
511 An example event packet context type:
513 struct event_packet_context {
514 uint64_t timestamp_begin;
515 uint64_t timestamp_end;
517 uint32_t stream_packet_count;
518 uint32_t events_discarded;
520 uint32_t/uint16_t content_size;
521 uint32_t/uint16_t packet_size;
522 uint8_t stream_packet_count_bits; /* Significant counter bits */
523 uint8_t compression_scheme;
524 uint8_t encryption_scheme;
530 The overall structure of an event is:
532 - Event Header (as specifed by the stream metadata)
533 - Extended Event Header (as specified by the event header)
534 - Event Context (as specified by the stream metadata)
535 - Event Payload (as specified by the event metadata)
540 One major factor can vary between streams: the number of event IDs assigned to
541 a stream. Luckily, this information tends to stay relatively constant (modulo
542 event registration while trace is being recorded), so we can specify different
543 representations for streams containing few event IDs and streams containing
544 many event IDs, so we end up representing the event ID and timestamp as densely
545 as possible in each case.
547 We therefore provide two types of events headers. Type 1 accommodates streams
548 with less than 31 event IDs. Type 2 accommodates streams with 31 or more event
551 The "extended headers" are used in the rare occasions where the information
552 cannot be represented in the ranges available in the event header. They are also
553 used in the rare occasions where the data required for a field could not be
554 collected: the flag corresponding to the missing field within the missing_fields
555 array is then set to 1.
557 Types uintX_t represent an X-bit unsigned integer.
560 6.1.1 Type 1 - Few event IDs
562 - Aligned on 32-bit (or 8-bit if byte-packed, depending on the architecture
564 - Fixed size: 32 bits.
565 - Native architecture byte ordering.
567 struct event_header_1 {
570 * id 31 is reserved to indicate a following
576 The end of a type 1 header is aligned on a 32-bit boundary (or packed).
579 6.1.2 Extended Type 1 Event Header
581 - Follows struct event_header_1, which is aligned on 32-bit, so no need to
583 - Variable size (depends on the number of fields per event).
584 - Native architecture byte ordering.
585 - NR_FIELDS is the number of fields within the event.
587 struct event_header_1_ext {
588 uint32_t id; /* 32-bit event IDs */
589 uint64_t timestamp; /* 64-bit timestamps */
590 uint1_t missing_fields[NR_FIELDS]; /* missing event fields bitmap */
594 6.1.3 Type 2 - Many event IDs
596 - Aligned on 32-bit (or 8-bit if byte-packed, depending on the architecture
598 - Fixed size: 48 bits.
599 - Native architecture byte ordering.
601 struct event_header_2 {
604 * id: range: 0 - 65534.
605 * id 65535 is reserved to indicate a following
610 The end of a type 2 header is aligned on a 16-bit boundary (or 8-bit if
614 6.1.4 Extended Type 2 Event Header
616 - Follows struct event_header_2, which alignment end on a 16-bit boundary, so
617 we need to align on 64-bit integer architecture alignment (or 8-bit if
619 - Variable size (depends on the number of fields per event).
620 - Native architecture byte ordering.
621 - NR_FIELDS is the number of fields within the event.
623 struct event_header_2_ext {
624 uint64_t timestamp; /* 64-bit timestamps */
625 uint32_t id; /* 32-bit event IDs */
626 uint1_t missing_fields[NR_FIELDS]; /* missing event fields bitmap */
632 The event context contains information relative to the current event. The choice
633 and meaning of this information is specified by the metadata "stream"
634 information. For this trace format, event context is usually empty, except when
635 the metadata "stream" information specifies otherwise by declaring a non-empty
636 structure for the event context. An example of event context is to save the
637 event payload size with each event, or to save the current PID with each event.
638 These are declared within the stream declaration within the metadata.
640 An example event context type:
642 struct event_context {
644 uint16_t payload_size;
650 An event payload contains fields specific to a given event type. The fields
651 belonging to an event type are described in the event-specific metadata
652 within a structure type.
656 No padding at the end of the event payload. This differs from the ISO/C standard
657 for structures, but follows the CTF standard for structures. In a trace, even
658 though it makes sense to align the beginning of a structure, it really makes no
659 sense to add padding at the end of the structure, because structures are usually
660 not followed by a structure of the same type.
662 This trick can be done by adding a zero-length "end" field at the end of the C
663 structures, and by using the offset of this field rather than using sizeof()
664 when calculating the size of a structure (see Appendix "A. Helper macros").
668 The event payload is aligned on the largest alignment required by types
669 contained within the payload. (This follows the ISO/C standard for structures)
675 The meta-data is located in a stream named "metadata". It is made of "event
676 packets", which each start with an event packet header. The event type within
677 the metadata stream have no event header nor event context. Each event only
678 contains a null-terminated "string" payload, which is a metadata description
679 entry. The events are packed one next to another. Each event packet start with
680 an event packet header, which contains, amongst other fields, the magic number
683 The metadata can be parsed by reading through the metadata strings, skipping
684 newlines and null-characters. Type names may contain spaces.
687 major = value; /* Trace format version */
689 uuid = value; /* Trace UUID */
696 /* Type 1 - Few event IDs; Type 2 - Many event IDs. See section 6.1. */
697 header_type = event_header_1 OR event_header_2;
699 * Extended event header type. Only present if specified in event header
700 * on a per-event basis.
702 header_type_ext = event_header_1_ext OR event_header_2_ext;
703 context_type = struct {
708 context_type = struct {
716 id = value; /* Numeric identifier within the stream */
723 /* More detail on types in section 4. Types */
726 typedef some existing type new_type;
740 /* Unnamed types, contained within compound type fields or type assignments. */
759 The two following macros keep track of the size of a GNU/C structure without
760 padding at the end by placing HEADER_END as the last field. A one byte end field
761 is used for C90 compatibility (C99 flexible arrays could be used here). Note
762 that this does not affect the effective structure size, which should always be
763 calculated with the header_sizeof() helper.
765 #define HEADER_END char end_field
766 #define header_sizeof(type) offsetof(typeof(type), end_field)
769 B. Stream Header Rationale
771 An event stream is divided in contiguous event packets of variable size. These
772 subdivisions allow the trace analyzer to perform a fast binary search by time
773 within the stream (typically requiring to index only the event packet headers)
774 without reading the whole stream. These subdivisions have a variable size to
775 eliminate the need to transfer the event packet padding when partially filled
776 event packets must be sent when streaming a trace for live viewing/analysis.
777 An event packet can contain a certain amount of padding at the end. Dividing
778 streams into event packets is also useful for network streaming over UDP and
779 flight recorder mode tracing (a whole event packet can be swapped out of the
780 buffer atomically for reading).
782 The stream header is repeated at the beginning of each event packet to allow
783 flexibility in terms of:
786 - allowing arbitrary buffers to be discarded without making the trace
788 - allow UDP packet loss handling by either dealing with missing event packet
789 or asking for re-transmission.
790 - transparently support flight recorder mode,
791 - transparently support crash dump.
793 The event stream header will therefore be referred to as the "event packet
794 header" throughout the rest of this document.