Skip to contentSkip to conversation

Add perf.data tracepoint events to trace.dat conversion

RFC PATCH Superseded Tanushree Shah 2 participants 10 messages Aug 2 – Aug 5 perf events

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover Add perf.data tracepoint events to trace.dat conversion

This RFC patch series introduces support for converting perf.data files containing tracepoint events into trace.dat format, enabling seamless visualization and analysis using KernelShark.

======================
Background and Motivation
======================

Currently, perf and trace-cmd operate as separate tracing ecosystems with incompatible data formats. Users who collect tracepoint data with 'perf record' cannot easily visualize it in KernelShark's graphical timeline view or leverage trace-cmd's analysis capabilities.

This creates workflow friction when users need to:

- Visualize perf tracepoint data in KernelShark's interactive graphical timeline
- Share trace data between perf and trace-cmd workflows and toolchains
- Perform architecture-independent conversion and analysis of traces

This conversion bridge eliminates these barriers by enabling seamless data exchange between perf and trace-cmd ecosystems, allowing users to choose the best tool for each analysis phase.

======================
Implementation Overview
======================

The series implements the trace.dat file format specification (version 7) within perf's data conversion framework.

**Patch 1/5: Core trace.dat Export Infrastructure** Introduces util/trace-dat.c and util/trace-dat.h implementing:
- Per-CPU raw event buffer management (init, collect, free)
- Ftrace ring buffer page construction
- trace.dat section writers (strings, options, flyrecord sections)

**Patch 2/5: Metadata Integration**
Extends util/trace-event-read.c to write trace.dat metadata during perf.data
parsing:
- Initial format header (magic, version, endian, page size, compression)
- Section 16: HEADER INFO (header_page + header_event)
- Section 17: FTRACE EVENT FORMATS
- Section 18: EVENT FORMATS (per system/event format files)
- Section 19: KALLSYMS
- Section 21: CMDLINES
- Section 15: STRINGS (written last after all sections)

**Patch 3/5: Conversion Backend**
Implements util/data-convert-trace.c with trace_convert__perf2dat() function:
- Processes PERF_TYPE_TRACEPOINT samples via process_sample_event()
- Collects raw event data per-CPU using trace_dat__collect_cpu_event()
- Writes OPTIONS sections (CPUCOUNT, TRACECLOCK, metadata offsets)
- Writes FLYRECORD section with per-CPU ring buffer pages

**Patch 4/5: User Interface**
Extends tools/perf/builtin-data.c with --to-trace-dat option:
- Adds command-line option for trace.dat output
- Mutually exclusive with --to-ctf and --to-json
- Calls trace_convert__perf2dat() to perform conversion

**Patch 5/5: Shell Test**
Adds a shell test (tools/perf/tests/shell/) covering normal tracepoint recordings, pipe mode and mixed tracepoint/non-tracepoint recording conversions and --force flag behaviour.

======================
Current Implementation Details
======================

**trace.dat Format Version:**
The implementation currently targets trace.dat format version 7, which is the stable version supported by current trace-cmd releases (v3.x). This version is hardcoded to ensure compatibility with existing trace-cmd and KernelShark installations. Future enhancements could add version negotiation or support for newer format versions as they become standardized.

**Compression Strategy:**
Compression is explicitly disabled (set to NONE) in the generated trace.dat files.
This design choice:
- Simplifies the initial implementation and testing
- Ensures maximum compatibility across trace-cmd versions
- Avoids external compression library dependencies

Future work could add support for various compression algorithms (zlib, zstd, lz4) with runtime selection via command-line options, significantly reducing file sizes for large traces.

======================
Usage Example
======================

```bash
*Record tracepoint events with perf*
perf record -e sched:sched_switch -e sched:sched_wakeup -a sleep 10

*Convert to trace.dat format*
perf data convert --to-trace-dat=output.dat

*Verify trace.dat structure*
trace-cmd dump --summary output.dat

*Analyze with trace-cmd*
trace-cmd report output.dat

*Visualize in KernelShark*
kernelshark output.dat
```

**Conversion Output:**

```
[ perf data convert: Converted 'perf.data' into trace.dat format
'output.dat' ]
[ perf data convert: Converted 2684 events ]
```
**trace-cmd dump --summary Output:**
```
 Tracing meta data in file output.dat:
	[Initial format]
		7	[Version]
		0	[Little endian]
		8	[Bytes in a long]
		65536	[Page size, bytes]
		none	[Compression algorithm]
			[Compression version]
	[buffer "", "local" clock, 65536 page size, 16 cpus, 1048576 bytes
    flyrecord data]
	[10 options]
	[Saved command lines, 0 bytes]
	[Kallsyms, 0 bytes]
	[Ftrace format, 0 events]
	[Header page, 206 bytes]
	[Header event, 205 bytes]
	[Events format, 1 systems]
	[9 sections]
```	

======================
Testing and Verification
======================

The series has been extensively tested with:
- Various tracepoint events (sched, irq, syscalls, block I/O)
- Mixed recordings containing both tracepoint and non-tracepoint events only tracepoints converted)
- Verification with trace-cmd report and KernelShark visualization
- Memory leak testing with Valgrind (0 bytes leaked).
- Cross-architecture testing: v1 tested x86_64 and ppc64le. v2 adds s390 (big-endian) perf.data converted on both ppc64le (little-endian) and x86_64 (little-endian) hosts, in addition to same-arch x86_64 (LE->LE) and ppc64le (BE->BE) conversion.
- Pipe mode support has been tested end-to-end. (in v2)

All generated trace.dat files successfully open in:
- trace-cmd report (v3.1+)
- KernelShark (v2.0+)

======================
Next Steps
======================

We would highly appreciate reviews, comments, and feedback on:
- The overall architectural approach and integration points
- Compatibility considerations with trace-cmd ecosystem
- Performance characteristics for large-scale traces
- Additional use cases or workflow scenarios
- Future enhancement priorities

---
Changes in v2

Addressing the Sashiko AI review findings on v1:

Cross-arch correctness:
- Introduce to_file_u16/u32/u64 helpers (wrapping tep_read_number()) to write all multi-byte fields in the recorded machine's byte order; apply throughout metadata sections and flyrecord page/record headers (ts, commit, TIME_EXTEND, large-event data_len).
- Fix flyrecord record header bit layout for big-endian files. The record header word bit layout differs by file endianness, matching kbuffer-parse.c type_len4host()/ts4host():
LE: type_len in bits [4:0], time_delta in bits [31:5]
BE: type_len in bits [31:27], time_delta in bits [26:0]

Pipe mode:
- Add process_attr(), process_feature(), and process_tracing_data() callbacks required for pipe mode operation.
- Defer CPU buffer initialisation until the first tracepoint sample, after process_feature()/process_tracing_data() have populated the session header. This ensures the recorded machine's CPU count is used rather than the host's - critical for cross-platform analysis.

Format compliance:
- Implement TIME_EXTEND records for timestamp deltas >27 bits to prevent silent truncation and maintain chronological ordering.
- Fix large event encoding (>=29 words): use type_len=0 with a separate 32-bit length word, avoiding collision with reserved types (PADDING=29, TIME_EXTEND=30, TIME_STAMP=31).
- Add bounds check rejecting records larger than a page payload before batching, preventing heap overflow in trace_dat__write_page().
- Fix flyrecord section_size to exclude the 16-byte section header, matching trace.dat specification and trace-cmd behaviour.

CLI behavior:
- Fix --force flag: open with O_CREAT|O_EXCL when force is not set, failing with -EEXIST instead of silently overwriting existing files.

Memory safety:
- Fix realloc overwrite of cpu_events->events and page_records on
failure: use temporary pointers, only commit on success.
- Fix use-after-free/double-free in sequential page_records realloc
failure: replace with malloc+memcpy+free pattern.
- Fix section_size computed from before section header position.
- Add NULL checks for get_tracing_file(), calloc() padding, and trace_dat_options_offset assignment on write failure.
- Use goto out_free on record allocation failure to avoid leaking accumulated page_records entries.
- Replace direct read() with do_read() in read_proc_kallsyms() to handle short reads correctly.
- On fwrite failure, set trace_dat_write_failed and continue parsing so that perf.data processing completes normally.

Testing (new in v2):
- Add shell test covering conversion, trace-cmd dump validation, sched_switch event verification, and --force flag behaviour.

Documentation (new in v2):
- Add documentation for 'perf data convert --to-trace-dat', covering usage and supported options.

v1: https://lore.kernel.org/linux-perf-users/[email protected]/

Tanushree Shah (5):
perf/trace-dat: Add trace.dat export infrastructure perf/trace-event: Write trace.dat metadata sections during parsing perf data-convert: Add perf.data to trace.dat conversion backend perf data: Add --to-trace-dat option for converting perf.data
tracepoint events into trace.dat format
perf test: Add test validating trace.dat generated by 'perf data

    convert --to-trace-dat'
 tools/perf/Documentation/perf-data.txt        |   7 +
 tools/perf/builtin-data.c                     |  40 +-
 ...rf_data_converter_tracepoints_trace_dat.sh | 169 ++++
 tools/perf/util/Build                         |   2 +
 tools/perf/util/data-convert-trace.c          | 240 +++++
 tools/perf/util/data-convert.h                |   4 +
 tools/perf/util/trace-dat.c                   | 876 ++++++++++++++++++
 tools/perf/util/trace-dat.h                   | 113 +++
 tools/perf/util/trace-event-read.c            | 307 +++++-
 9 files changed, 1749 insertions(+), 9 deletions(-)
 create mode 100755 tools/perf/tests/shell/test_perf_data_converter_tracepoints_trace_dat.sh
 create mode 100644 tools/perf/util/data-convert-trace.c
 create mode 100644 tools/perf/util/trace-dat.c
 create mode 100644 tools/perf/util/trace-dat.h
Signature · 1 line
2.47.3
#1 lore raw

RFC PATCH v2 1/5 perf/trace-dat: Add trace.dat export infrastructure

Add new utility files util/trace-dat.c and util/trace-dat.h implementing the infrastructure for exporting perf.data tracepoints to trace.dat format compatible with trace-cmd and KernelShark.

trace-dat.c defines all globals and functions needed for:
- Per-cpu raw event buffer management (init_cpu_buffers, collect_cpu_event, free_cpu_buffers).
- ftrace ring buffer page construction (write_page, write_cpu_dat), including TIME_EXTEND records for large timestamp deltas.
- trace.dat section writers (write_strings_section, write_options_section1, write_options_section2, write_flyrecord_section).

trace-dat.h declares all globals and function prototypes to be used by data-convert-trace.c and trace-event-read.c.

---
 tools/perf/util/Build       |   1 +
 tools/perf/util/trace-dat.c | 825 ++++++++++++++++++++++++++++++++++++
 tools/perf/util/trace-dat.h |  83 ++++
 3 files changed, 909 insertions(+)
 create mode 100644 tools/perf/util/trace-dat.c
 create mode 100644 tools/perf/util/trace-dat.h
tools/perf/util/Build +1 -0
index 70cc91d00804..c000d8032d25 100644
@@ -98,6 +98,7 @@ perf-util-y += trace-event-scripting.o
 perf-util-$(CONFIG_LIBTRACEEVENT) += trace-event.o
 perf-util-$(CONFIG_LIBTRACEEVENT) += trace-event-parse.o
 perf-util-$(CONFIG_LIBTRACEEVENT) += trace-event-read.o
+perf-util-$(CONFIG_LIBTRACEEVENT) += trace-dat.o
 perf-util-y += sort.o
 perf-util-y += hist.o
 perf-util-y += util.o
tools/perf/util/trace-dat.c new file +825 -0

825 lines not on this page · raw message

tools/perf/util/trace-dat.h new file +83 -0

83 lines not on this page · raw message

Signature · 1 line
2.47.3
#2 lore raw

RFC PATCH v2 2/5 perf/trace-event: Write trace.dat metadata sections during parsing

In reply to Tanushree Shah

Perf already captures the tracing metadata as a part of data section in perf.data

When trace_dat_fp is set, write trace.dat compatible metadata sections using the perf provided raw buffers.

Sections written:
- Initial format header (magic, version, endian, long_size, page_size, compression, options_offset placeholder)
- Section 16: HEADER INFO (header_page + header_event)
- Section 17: FTRACE EVENT FORMATS
- Section 18: EVENT FORMATS (per system/event format files)
- Section 19: KALLSYMS
- Section 21: CMDLINES
- Section 15: STRINGS (written last after all sections)

Introduce to_file_u16/u32/u64 helpers wrapping tep_read_number() to write all trace.dat metadata in the recorded machine's byte order, ensuring correct cross-platform compatibility.

---
 tools/perf/util/trace-dat.c        | 212 ++++++++++++--------
 tools/perf/util/trace-dat.h        |  40 +++-
 tools/perf/util/trace-event-read.c | 307 ++++++++++++++++++++++++++++-
 3 files changed, 465 insertions(+), 94 deletions(-)
tools/perf/util/trace-dat.c +130 -82
index 55a7bd982c6f..f71e03716e27 100644
@@ -43,6 +43,7 @@
 /* Buffer size for reading trace_clock string from debugfs/tracefs */
 #define CLOCK_BUFFER_SIZE 256
 
+bool trace_dat_write_failed;
 FILE *trace_dat_fp;
 int trace_dat_page_size;
 int trace_dat_nr_cpus;
@@ -143,10 +144,12 @@ int trace_dat__collect_cpu_event(int cpu, unsigned long long ts,
 }
 
 /* Write a single page of trace records */
-static int trace_dat__write_page(FILE *fp, unsigned long long base_ts,
+static int trace_dat__write_page(FILE *fp, struct tep_handle *pevent, unsigned long long base_ts,
 			char **records, int *rec_sizes, int nr_recs)
 {
 	unsigned long long commit = 0;
+	unsigned long long ts_out;
+	unsigned long long commit_out;
 	int offset = TRACE_DAT_RECORD_HEADER_SIZE;
 	int i;
 	char *page;
@@ -161,8 +164,12 @@ static int trace_dat__write_page(FILE *fp, unsigned long long base_ts,
 		commit += rec_sizes[i];
 	}
 
-	memcpy(page, &base_ts, sizeof(base_ts));
-	memcpy(page + sizeof(base_ts), &commit, sizeof(commit));
+	/* Byte-swap page header for cross-arch compatibility */
+	ts_out = to_file_u64(pevent, base_ts);
+	commit_out = to_file_u64(pevent, commit);
+
+	memcpy(page, &ts_out, sizeof(ts_out));
+	memcpy(page + sizeof(ts_out), &commit_out, sizeof(commit_out));
 
 	if (!fwrite(page, 1, trace_dat_page_size, fp)) {
 		free(page);
@@ -174,7 +181,8 @@ static int trace_dat__write_page(FILE *fp, unsigned long long base_ts,
 }
 
 /* Write all trace data for a single CPU as trace.dat flyrecord pages */
-static int trace_dat__write_cpu_dat(FILE *fp, int cpu, unsigned long long *file_offset_out)
+static int trace_dat__write_cpu_dat(FILE *fp, struct tep_handle *pevent,
+		int cpu, unsigned long long *file_offset_out)
 {
 	struct cpu_events *cpu_events = &trace_cpu_data[cpu];
 	unsigned long long base_ts;
@@ -229,11 +237,18 @@ static int trace_dat__write_cpu_dat(FILE *fp, int cpu, unsigned long long *file_
 			if (!extend)
 				return -ENOMEM;
 
-			extend_hdr =
-				((time_delta & TRACE_DAT_RECORD_TIME_MASK) <<
-				 TRACE_DAT_RECORD_TIME_SHIFT) |
-				TRACE_DAT_RECORD_TYPE_TIME_EXTEND;
-			delta_upper = time_delta >> TRACE_DAT_RECORD_TIME_SHIFT;
+			if (tep_is_file_bigendian(pevent)) {
+				extend_hdr = (time_delta & TRACE_DAT_RECORD_TIME_MASK) |
+				     (TRACE_DAT_RECORD_TYPE_TIME_EXTEND << 27);
+				delta_upper = time_delta >> 27;
+			} else {
+				extend_hdr = ((time_delta & TRACE_DAT_RECORD_TIME_MASK) <<
+					TRACE_DAT_RECORD_TIME_SHIFT) |
+					TRACE_DAT_RECORD_TYPE_TIME_EXTEND;
+				delta_upper = time_delta >> TRACE_DAT_RECORD_TIME_SHIFT;
+			}
+			extend_hdr  = to_file_u32(pevent, extend_hdr);   /* still needed */
+			delta_upper = to_file_u32(pevent, delta_upper);   /* still needed */
 
 			memcpy(extend, &extend_hdr, TRACE_DAT_WORD_SIZE);
 			memcpy(extend + TRACE_DAT_WORD_SIZE, &delta_upper,
@@ -273,7 +288,7 @@ static int trace_dat__write_cpu_dat(FILE *fp, int cpu, unsigned long long *file_
 		/* Check page fit BEFORE allocating data record */
 		if (page_size_used + needed_size >
 			trace_dat_page_size - TRACE_DAT_RECORD_HEADER_SIZE) {
-			ret = trace_dat__write_page(fp, base_ts,
+			ret = trace_dat__write_page(fp, pevent, base_ts,
 					page_records, page_rec_sizes,
 					nr_page_recs);
 
@@ -299,7 +314,13 @@ static int trace_dat__write_cpu_dat(FILE *fp, int cpu, unsigned long long *file_
 			time_delta = 0;
 		}
 
-		hdr_word = (time_delta << TRACE_DAT_RECORD_TIME_SHIFT) | type_len;
+		if (tep_is_file_bigendian(pevent))
+			hdr_word = ((unsigned int)time_delta & TRACE_DAT_RECORD_TIME_MASK) |
+				(type_len << 27);
+		else
+			hdr_word = ((unsigned int)time_delta << TRACE_DAT_RECORD_TIME_SHIFT) |
+				type_len;
+		hdr_word = to_file_u32(pevent, hdr_word);
 
 		data_rec = calloc(1, data_rec_size);
 		if (!data_rec) {
@@ -311,8 +332,11 @@ static int trace_dat__write_cpu_dat(FILE *fp, int cpu, unsigned long long *file_
 		memcpy(data_rec, &hdr_word, TRACE_DAT_WORD_SIZE);
 
 		/* Large events: write actual byte length after header */
-		if (type_len == 0)
-			memcpy(data_rec + TRACE_DAT_WORD_SIZE, &data_len, TRACE_DAT_WORD_SIZE);
+		if (type_len == 0) {
+			unsigned int data_len_out = to_file_u32(pevent, data_len);
+
+			memcpy(data_rec + TRACE_DAT_WORD_SIZE, &data_len_out, TRACE_DAT_WORD_SIZE);
+		}
 
 		memcpy(data_rec + payload_offset, event->raw, data_len);
 
@@ -366,7 +390,7 @@ static int trace_dat__write_cpu_dat(FILE *fp, int cpu, unsigned long long *file_
 	}
 
 	if (nr_page_recs > 0) {
-		ret = trace_dat__write_page(fp, base_ts,
+		ret = trace_dat__write_page(fp, pevent, base_ts,
 				page_records, page_rec_sizes, nr_page_recs);
 	}
 out_free:
@@ -378,11 +402,11 @@ static int trace_dat__write_cpu_dat(FILE *fp, int cpu, unsigned long long *file_
 }
 
 /* Write the strings section containing section name lookup table */
-int trace_dat__write_strings_section(void)
+int trace_dat__write_strings_section(struct tep_handle *pevent)
 {
-	unsigned short section_id = TRACE_DAT_SECTION_STRINGS;
-	unsigned short flags = 0;
 	unsigned long long section_size = 0;
+	unsigned short section_id = to_file_u16(pevent, TRACE_DAT_SECTION_STRINGS);
+	unsigned short flags = to_file_u16(pevent, 0);
 	static const char * const section_names[] = {
 		"headers",		/* offset 0 - strid for section 16  */
 		"ftrace event formats", /* offset 8 - strid for section 17  */
@@ -397,7 +421,7 @@ int trace_dat__write_strings_section(void)
 	};
 
 	/* string_id points to "strings" string itself */
-	unsigned int string_id = STRID_STRINGS;
+	unsigned int string_id = to_file_u32(pevent, STRID_STRINGS);
 	int i;
 
 	if (!trace_dat_fp)
@@ -405,47 +429,59 @@ int trace_dat__write_strings_section(void)
 
 	for (i = 0; section_names[i] != NULL; i++)
 		section_size += strlen(section_names[i]) + 1;
+	section_size = to_file_u64(pevent, section_size);
 
 	/* write section header */
 	if (!fwrite(&section_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 		       !fwrite(&flags, sizeof(unsigned short), 1, trace_dat_fp) ||
 		       !fwrite(&string_id, sizeof(unsigned int), 1, trace_dat_fp) ||
-		       !fwrite(&section_size, sizeof(unsigned long long), 1, trace_dat_fp))
-		return -EIO;
+		       !fwrite(&section_size, sizeof(unsigned long long), 1, trace_dat_fp)) {
+		pr_warning("Failed to write strings section\n");
+		trace_dat_write_failed = true;
+	}
 
 	/* write strings */
 	for (i = 0; section_names[i] != NULL; i++)
-		if (!fwrite(section_names[i], 1, strlen(section_names[i]) + 1, trace_dat_fp))
-			return -EIO;
+		if (!fwrite(section_names[i], 1, strlen(section_names[i]) + 1, trace_dat_fp)) {
+			pr_warning("Failed to write strings section\n");
+			trace_dat_write_failed = true;
+		}
 	return 0;
 }
 
 /* Writes options section containing CPUCOUNT, TRACECLOCK, EVENT_FORMAT, HEADER_INFO,
  * FTRACE_EVENTS, KALLSYMS, CMDLINES options, ending with DONE option pointing to next section.
  */
-int trace_dat__write_options_section1(void)
+int trace_dat__write_options_section1(struct tep_handle *pevent)
 {
-	unsigned short section_id = TRACE_DAT_SECTION_OPTIONS;
-	unsigned short flags = 0;
-	unsigned int string_id = STRID_OPTIONS_1;
+	unsigned short section_id = to_file_u16(pevent, TRACE_DAT_SECTION_OPTIONS);
+	unsigned short flags = to_file_u16(pevent, 0);
+	unsigned int string_id = to_file_u32(pevent, STRID_OPTIONS_1);
 	unsigned long long section_size = 0;
 	long section_size_pos;
 	long payload_start;
 	unsigned long long section_start;
 	unsigned short opt_id;
 	unsigned int opt_size;
+	unsigned int raw_opt_size;
+	unsigned int nr_cpus;
 	char clock_buf[CLOCK_BUFFER_SIZE];
 	FILE *clock_file;
 	size_t bytes_read;
 	char *path;
 	unsigned long long next_offset;
+	unsigned long long events_off;
+	unsigned long long header_off;
+	unsigned long long ftrace_off;
+	unsigned long long kallsyms_off;
+	unsigned long long cmdline_off;
 	long end_pos;
 
 	if (!trace_dat_fp)
 		return -EBADF;
 
 	/* fill options_offset in initial format */
-	section_start = ftell(trace_dat_fp);
+	section_start = to_file_u64(pevent, ftell(trace_dat_fp));
 
 	if (fseek(trace_dat_fp, trace_dat_options_offset, SEEK_SET) < 0 ||
 	   !fwrite(&section_start, sizeof(unsigned long long), 1, trace_dat_fp) ||
@@ -464,16 +500,17 @@ int trace_dat__write_options_section1(void)
 	payload_start = ftell(trace_dat_fp);
 
 	/* CPUCOUNT option */
-	opt_id = TRACE_DAT_OPTION_CPUCOUNT;
-	opt_size = sizeof(unsigned int);
+	opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_CPUCOUNT);
+	opt_size = to_file_u32(pevent, sizeof(unsigned int));
+	nr_cpus = to_file_u32(pevent, trace_dat_nr_cpus);
 
 	if (!fwrite(&opt_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 	    !fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp) ||
-	    !fwrite(&trace_dat_nr_cpus, sizeof(unsigned int), 1, trace_dat_fp))
+	    !fwrite(&nr_cpus, sizeof(unsigned int), 1, trace_dat_fp))
 		return -EIO;
 
 	/* TRACECLOCK option */
-	opt_id = TRACE_DAT_OPTION_TRACECLOCK;
+	opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_TRACECLOCK);
 
 	path = get_tracing_file("trace_clock");
 	if (path) {
@@ -490,66 +527,72 @@ int trace_dat__write_options_section1(void)
 		strcpy(clock_buf, "local\n");
 		bytes_read = strlen(clock_buf);
 	}
-	opt_size = bytes_read + 1;
+	raw_opt_size = bytes_read + 1;
+	opt_size = to_file_u32(pevent, bytes_read + 1);
 	if (!fwrite(&opt_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 	    !fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp) ||
-	    !fwrite(clock_buf, 1, opt_size, trace_dat_fp))
+	    !fwrite(clock_buf, 1, raw_opt_size, trace_dat_fp))
 		return -EIO;
 
 	/* EVENT option */
-	opt_id = TRACE_DAT_OPTION_EVENT;
-	opt_size = sizeof(unsigned long long);
+	opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_EVENT);
+	opt_size = to_file_u32(pevent, sizeof(unsigned long long));
+	events_off = to_file_u64(pevent, trace_dat_events_format_offset);
 
 	if (!fwrite(&opt_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 	   !fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp) ||
-	   !fwrite(&trace_dat_events_format_offset, sizeof(unsigned long long),
+	   !fwrite(&events_off, sizeof(unsigned long long),
 		   1, trace_dat_fp))
 		return -EIO;
 
 	/* HEADER option */
-	opt_id = TRACE_DAT_OPTION_HEADER;
-	opt_size = sizeof(unsigned long long);
+	opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_HEADER);
+	opt_size = to_file_u32(pevent, sizeof(unsigned long long));
+	header_off = to_file_u64(pevent, trace_dat_header_info_offset);
 
 	if (!fwrite(&opt_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 	    !fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp) ||
-	    !fwrite(&trace_dat_header_info_offset, sizeof(unsigned long long),
+	    !fwrite(&header_off, sizeof(unsigned long long),
 		    1, trace_dat_fp))
 		return -EIO;
 
 	/* FTRACE option */
-	opt_id = TRACE_DAT_OPTION_FTRACE;
-	opt_size = sizeof(unsigned long long);
+	opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_FTRACE);
+	opt_size = to_file_u32(pevent, sizeof(unsigned long long));
+	ftrace_off = to_file_u64(pevent, trace_dat_ftrace_format_offset);
 
 	if (!fwrite(&opt_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 	    !fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp) ||
-	    !fwrite(&trace_dat_ftrace_format_offset, sizeof(unsigned long long),
+	    !fwrite(&ftrace_off, sizeof(unsigned long long),
 		   1, trace_dat_fp))
 		return -EIO;
 
 	/* KALLSYMS option */
-	opt_id = TRACE_DAT_OPTION_KALLSYMS;
-	opt_size = sizeof(unsigned long long);
+	opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_KALLSYMS);
+	opt_size = to_file_u32(pevent, sizeof(unsigned long long));
+	kallsyms_off = to_file_u64(pevent, trace_dat_kallsyms_offset);
 
 	if (!fwrite(&opt_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 	    !fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp) ||
-	    !fwrite(&trace_dat_kallsyms_offset, sizeof(unsigned long long),
+	    !fwrite(&kallsyms_off, sizeof(unsigned long long),
 		    1, trace_dat_fp))
 		return -EIO;
 
 	/* CMDLINE option */
-	opt_id = TRACE_DAT_OPTION_CMDLINE;
-	opt_size = sizeof(unsigned long long);
+	opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_CMDLINE);
+	opt_size = to_file_u32(pevent, sizeof(unsigned long long));
+	cmdline_off = to_file_u64(pevent, trace_dat_cmdline_offset);
 
 	if (!fwrite(&opt_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 	    !fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp) ||
-	    !fwrite(&trace_dat_cmdline_offset, sizeof(unsigned long long),
+	    !fwrite(&cmdline_off, sizeof(unsigned long long),
 		    1, trace_dat_fp))
 		return -EIO;
 
 	/* DONE option id - next_options_offset filled later */
-	opt_id = TRACE_DAT_OPTION_DONE;
-	opt_size = sizeof(unsigned long long);
-	next_offset = 0;  /* placeholder */
+	opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_DONE);
+	opt_size = to_file_u32(pevent, sizeof(unsigned long long));
+	next_offset = to_file_u64(pevent, 0);
 
 	trace_dat_next_options_offset = ftell(trace_dat_fp);
 	if (!fwrite(&opt_id, sizeof(unsigned short), 1, trace_dat_fp) ||
@@ -560,7 +603,7 @@ int trace_dat__write_options_section1(void)
 	/* fill section size */
 	end_pos = ftell(trace_dat_fp);
 
-	section_size = end_pos - payload_start;
+	section_size = to_file_u64(pevent, end_pos - payload_start);
 	if (fseek(trace_dat_fp, section_size_pos, SEEK_SET) < 0 ||
 	    !fwrite(&section_size, sizeof(unsigned long long), 1, trace_dat_fp) ||
 	    fseek(trace_dat_fp, end_pos, SEEK_SET) < 0)
@@ -574,20 +617,21 @@ int trace_dat__write_options_section1(void)
  * (flyrecord section offset, clock type, page size, CPU count,
  * per-CPU offsets/sizes) and DONE option.
  */
-int trace_dat__write_options_section2(void)
+int trace_dat__write_options_section2(struct tep_handle *pevent)
 {
-	unsigned short section_id  = TRACE_DAT_SECTION_OPTIONS;
-	unsigned short flags = 0;
-	unsigned int string_id = STRID_OPTIONS_2;
+	unsigned short section_id = to_file_u16(pevent, TRACE_DAT_SECTION_OPTIONS);
+	unsigned short flags = to_file_u16(pevent, 0);
+	unsigned int string_id = to_file_u32(pevent, STRID_OPTIONS_2);
 	unsigned long long section_size = 0;
 	long section_size_pos;
 	long payload_start;
 	int cpu;
-	unsigned short opt_id = TRACE_DAT_OPTION_BUFFER;
-	unsigned int opt_size = 0;
+	unsigned short opt_id = to_file_u16(pevent, TRACE_DAT_OPTION_BUFFER);
+	unsigned int opt_size = 0;  /* filled after payload written */
 	long opt_size_pos;
 	unsigned long long data_offset = 0;
-	unsigned int page_size = (unsigned int)trace_dat_page_size;
+	unsigned int page_size = to_file_u32(pevent, (unsigned int)trace_dat_page_size);
+	unsigned int nr_cpus   = to_file_u32(pevent, trace_dat_nr_cpus);
 	const char *clock = "local";
 	unsigned long long next;
 	long end_pos;
@@ -600,7 +644,7 @@ int trace_dat__write_options_section2(void)
 		return -EINVAL;
 
 	/* fill done1 next offset - points to this section */
-	next = ftell(trace_dat_fp);
+	next = to_file_u64(pevent, ftell(trace_dat_fp));
 
 	if (fseek(trace_dat_fp, trace_dat_next_options_offset + 2 + 4, SEEK_SET) < 0 ||
 	    !fwrite(&next, sizeof(unsigned long long), 1, trace_dat_fp) ||
@@ -631,15 +675,16 @@ int trace_dat__write_options_section2(void)
 	    !fwrite("\0", 1, 1, trace_dat_fp) ||
 	    !fwrite(clock, 1, strlen(clock) + 1, trace_dat_fp) ||
 	    !fwrite(&page_size, sizeof(unsigned int), 1, trace_dat_fp) ||
-	    !fwrite(&trace_dat_nr_cpus, sizeof(unsigned int), 1, trace_dat_fp))
+	    !fwrite(&nr_cpus, sizeof(unsigned int), 1, trace_dat_fp))
 		return -EIO;
 
 	/* per cpu: cpu_id + offset placeholder + size */
 	for (cpu = 0; cpu < trace_dat_nr_cpus; cpu++) {
+		unsigned int cpu_out          = to_file_u32(pevent, cpu);
 		cpu_offset = 0;  /* filled in write_flyrecord */
-		cpu_size   = 0;  /* filled in write_flyrecord */
+		cpu_size = 0;  /* filled in write_flyrecord */
 
-		if (!fwrite(&cpu, sizeof(unsigned int), 1, trace_dat_fp))
+		if (!fwrite(&cpu_out, sizeof(unsigned int), 1, trace_dat_fp))
 			return -EIO;
 		buffer_opt_cpu_offsets_pos[cpu] = ftell(trace_dat_fp);
 		if (!fwrite(&cpu_offset, sizeof(unsigned long long), 1, trace_dat_fp) ||
@@ -650,17 +695,17 @@ int trace_dat__write_options_section2(void)
 	/* fill opt_size */
 	end_pos = ftell(trace_dat_fp);
 
-	opt_size = end_pos - opt_payload_start;
-	fseek(trace_dat_fp, opt_size_pos, SEEK_SET);
-	if (!fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp))
+	opt_size = to_file_u32(pevent, end_pos - opt_payload_start);
+	if (fseek(trace_dat_fp, opt_size_pos, SEEK_SET) < 0 ||
+	    !fwrite(&opt_size, sizeof(unsigned int), 1, trace_dat_fp) ||
+	    fseek(trace_dat_fp, end_pos, SEEK_SET) < 0)
 		return -EIO;
-	fseek(trace_dat_fp, end_pos, SEEK_SET);
 
 	/* DONE id=0 */
-	done_id = TRACE_DAT_OPTION_DONE;
-	done_size = sizeof(unsigned long long);
+	done_id = to_file_u16(pevent, TRACE_DAT_OPTION_DONE);
+	done_size = to_file_u32(pevent, sizeof(unsigned long long));
 	/* No additional options sections follow this one */
-	next = 0;
+	next = to_file_u64(pevent, 0);
 
 	if (!fwrite(&done_id, sizeof(unsigned short), 1, trace_dat_fp) ||
 	    !fwrite(&done_size, sizeof(unsigned int), 1, trace_dat_fp) ||
@@ -670,24 +715,25 @@ int trace_dat__write_options_section2(void)
 	/* fill section size */
 	end_pos = ftell(trace_dat_fp);
 
-	section_size = end_pos - payload_start;
-	fseek(trace_dat_fp, section_size_pos, SEEK_SET);
-	if (!fwrite(&section_size, sizeof(unsigned long long), 1, trace_dat_fp))
+	section_size = to_file_u64(pevent, end_pos - payload_start);
+	if (fseek(trace_dat_fp, section_size_pos, SEEK_SET) < 0 ||
+	    !fwrite(&section_size, sizeof(unsigned long long), 1, trace_dat_fp) ||
+	    fseek(trace_dat_fp, end_pos, SEEK_SET) < 0)
 		return -EIO;
-	fseek(trace_dat_fp, end_pos, SEEK_SET);
 
 	return 0;
 
 }
 
-int trace_dat__write_flyrecord_section(void)
+int trace_dat__write_flyrecord_section(struct tep_handle *pevent)
 {
-	unsigned short section_id = TRACE_DAT_SECTION_FLYRECORD;
-	unsigned short flags = 0;
-	unsigned int string_id = STRID_BUFFER_FLYRECORD;
+	unsigned short section_id = to_file_u16(pevent, TRACE_DAT_SECTION_FLYRECORD);
+	unsigned short flags = to_file_u16(pevent, 0);
+	unsigned int string_id = to_file_u32(pevent, STRID_BUFFER_FLYRECORD);
 	unsigned long long section_size = 0;
 	long section_size_pos;
 	long flyrecord_start;
+	unsigned long long flyrecord_out;
 	long after_header;
 	long padding_needed;
 	unsigned long long *cpu_offsets;
@@ -750,7 +796,7 @@ int trace_dat__write_flyrecord_section(void)
 	for (cpu = 0; cpu < trace_dat_nr_cpus; cpu++) {
 		start = ftell(trace_dat_fp);
 
-		ret = trace_dat__write_cpu_dat(trace_dat_fp, cpu, &cpu_offsets[cpu]);
+		ret = trace_dat__write_cpu_dat(trace_dat_fp, pevent, cpu, &cpu_offsets[cpu]);
 
 		if (ret < 0) {
 			pr_err("Failed to write CPU %d data\n", cpu);
@@ -762,7 +808,7 @@ int trace_dat__write_flyrecord_section(void)
 	/* fill section size */
 	end_pos = ftell(trace_dat_fp);
 
-	section_size = end_pos - after_header;
+	section_size = to_file_u64(pevent, end_pos - after_header);
 	if (fseek(trace_dat_fp, section_size_pos, SEEK_SET) < 0 ||
 	    !fwrite(&section_size, sizeof(unsigned long long), 1, trace_dat_fp)) {
 		ret = -EIO;
@@ -775,6 +821,8 @@ int trace_dat__write_flyrecord_section(void)
 
 	/* fill cpu offsets and sizes in BUFFER option */
 	for (cpu = 0; cpu < trace_dat_nr_cpus; cpu++) {
+		cpu_offsets[cpu] = to_file_u64(pevent, cpu_offsets[cpu]);
+		cpu_sizes[cpu]   = to_file_u64(pevent, cpu_sizes[cpu]);
 		if (fseek(trace_dat_fp, buffer_opt_cpu_offsets_pos[cpu], SEEK_SET) < 0 ||
 		    !fwrite(&cpu_offsets[cpu], sizeof(unsigned long long), 1, trace_dat_fp) ||
 		    !fwrite(&cpu_sizes[cpu], sizeof(unsigned long long), 1, trace_dat_fp)) {
@@ -784,8 +832,9 @@ int trace_dat__write_flyrecord_section(void)
 	}
 
 	/* fill data offset in buffer option */
+	flyrecord_out = to_file_u64(pevent, flyrecord_start);
 	if (fseek(trace_dat_fp, opt_payload_start, SEEK_SET) < 0 ||
-	    !fwrite(&flyrecord_start, sizeof(unsigned long long), 1, trace_dat_fp)) {
+	    !fwrite(&flyrecord_out, sizeof(unsigned long long), 1, trace_dat_fp)) {
 		ret = -EIO;
 		goto cleanup;
 	}
@@ -795,7 +844,6 @@ int trace_dat__write_flyrecord_section(void)
 		goto cleanup;
 	}
 
-
 cleanup:
 	free(cpu_offsets);
 	free(cpu_sizes);
tools/perf/util/trace-dat.h +35 -5
index 9aec37b708d4..5985083b275a 100644
@@ -8,9 +8,13 @@
 #define __PERF_TRACE_DAT_H
 
 #include <stdio.h>
+#include <stdbool.h>
+#include <event-parse.h>
+#include <byteswap.h>
+#include "util.h"
 
 /* trace.dat file format version */
-#define TRACE_DAT_VERSION '7'
+#define TRACE_DAT_VERSION "7"
 
 /*
  * Section IDs for trace.dat format
@@ -51,6 +55,32 @@
 #define STRID_OPTIONS_2        77
 #define STRID_BUFFER_FLYRECORD 85
 
+/*
+ * Set by trace.dat section writers (kallsyms, ftrace, events etc.) on
+ * fwrite() failure; checked by trace_convert__perf2dat() to clean up.
+ */
+extern bool trace_dat_write_failed;
+
+/*
+ * tep handle from the recording machine; set by trace_convert__perf2dat()
+ * before options/flyrecord sections are written.
+ */
+
+static inline uint16_t to_file_u16(struct tep_handle *pevent, uint16_t val)
+{
+	return tep_read_number(pevent, &val, 2);
+}
+
+static inline uint32_t to_file_u32(struct tep_handle *pevent, uint32_t val)
+{
+	return tep_read_number(pevent, &val, 4);
+}
+
+static inline uint64_t to_file_u64(struct tep_handle *pevent, uint64_t val)
+{
+	return tep_read_number(pevent, &val, 8);
+}
+
 struct perf_session;
 
 extern FILE *trace_dat_fp;
@@ -75,9 +105,9 @@ int trace_dat__collect_cpu_event(int cpu, unsigned long long ts,
 void trace_dat__free_cpu_buffers(void);
 
 /* write trace.dat file sections */
-int trace_dat__write_options_section1(void);
-int trace_dat__write_options_section2(void);
-int trace_dat__write_flyrecord_section(void);
-int trace_dat__write_strings_section(void);
+int trace_dat__write_options_section1(struct tep_handle *pevent);
+int trace_dat__write_options_section2(struct tep_handle *pevent);
+int trace_dat__write_flyrecord_section(struct tep_handle *pevent);
+int trace_dat__write_strings_section(struct tep_handle *pevent);
 
 #endif /* __PERF_TRACE_DAT_H */
tools/perf/util/trace-event-read.c +300 -7

448 lines not on this page · raw message

Signature · 1 line
2.47.3
#3 lore raw

RFC PATCH v2 3/5 perf data-convert: Add perf.data to trace.dat conversion backend

In reply to Tanushree Shah

Add data-convert-trace.c implementing trace_convert__perf2dat() to convert perf.data tracepoint events to trace.dat format.

process_sample_event() is invoked for each PERF_TYPE_TRACEPOINT sample during perf_session__process_events(), storing raw event bytes per-cpu via trace_dat__collect_cpu_event().

Once all samples are collected:
- trace_dat__write_options_section1() writes the OPTIONS section with CPUCOUNT, TRACECLOCK, HEADER_INFO, FTRACE_EVENTS, EVENT_FORMATS, KALLSYMS, CMDLINES and DONE options.
- trace_dat__write_options_section2() writes the OPTIONS section with BUFFER option holding per-cpu data offset placeholders and the DONE option.
- trace_dat__write_flyrecord_section() builds ring buffer pages per-cpu and patches BUFFER option with final offsets and sizes,

Per-cpu buffers are sized to tep_get_page_size() from the session tep handle and released on all exit paths.

Register .attr, .feature, and .tracing_data callbacks on the tool, delegating to the standard perf_event__process_attr(), perf_event__process_feature(), and perf_event__process_tracing_data() helpers, so these records are processed as they arrive in the stream instead of being left as default no-op stubs.

Defer per-CPU buffer initialisation to the first call of the .sample callback instead of doing it eagerly after session open. By this point, pipe-mode ordering guarantees (record__synthesize()) that attr, feature, and tracing_data records have already been written and processed, so nr_cpus_online and the tep page size are guaranteed to be populated from the recording machine's own stream data. This is required because the converting host may differ in architecture, CPU count, or page size from the machine that recorded the data, so these values must never be taken from local sysconf() or similar.

The output path honours --force: without it, the destination file is opened with O_CREAT|O_EXCL so an existing file is detected atomically and conversion fails with -EEXIST rather than silently overwriting prior output.

---
 tools/perf/util/Build                |   1 +
 tools/perf/util/data-convert-trace.c | 240 +++++++++++++++++++++++++++
 tools/perf/util/data-convert.h       |   4 +
 3 files changed, 245 insertions(+)
 create mode 100644 tools/perf/util/data-convert-trace.c
tools/perf/util/Build +1 -0
index c000d8032d25..88022b24e170 100644
@@ -236,6 +236,7 @@ ifeq ($(CONFIG_LIBTRACEEVENT),y)
 endif
 
 perf-util-y += data-convert-json.o
+perf-util-$(CONFIG_LIBTRACEEVENT) += data-convert-trace.o
 
 perf-util-y += scripting-engines/
 
tools/perf/util/data-convert-trace.c new file +240 -0

240 lines not on this page · raw message

tools/perf/util/data-convert.h +4 -0
index ee651fa680a1..d958e68367fe 100644
@@ -19,4 +19,8 @@ int bt_convert__perf2ctf(const char *input_name, const char *to_ctf,
 int bt_convert__perf2json(const char *input_name, const char *to_ctf,
 			 struct perf_data_convert_opts *opts);
 
+#ifdef HAVE_LIBTRACEEVENT
+int trace_convert__perf2dat(const char *input, const char *to_trace,
+			   struct perf_data_convert_opts *opts);
+#endif /* HAVE_LIBTRACEEVENT */
 #endif /* __DATA_CONVERT_H */
Signature · 1 line
2.47.3
#4 lore raw

RFC PATCH v2 4/5 perf data: Add --to-trace-dat option for converting perf.data tracepoint events into trace.dat format

In reply to Tanushree Shah

Add new command-line option to perf data convert for generating trace.dat output files.

The --to-trace-dat option:
- Accepts output filename for trace.dat format
- Mutually exclusive with --to-ctf and --to-json
- Calls trace_convert__perf2dat() to perform conversion

Usage:

  $ perf record -e sched:* -a sleep 1
  $ perf data convert --to-trace-dat=trace.dat
  $ trace-cmd report trace.dat

Document the new option in Documentation/perf-data.txt alongside the existing --to-ctf and --to-json entries.

---
 tools/perf/Documentation/perf-data.txt |  7 +++++
 tools/perf/builtin-data.c              | 40 ++++++++++++++++++++++++--
 tools/perf/util/trace-dat.c            |  3 ++
 3 files changed, 48 insertions(+), 2 deletions(-)
tools/perf/Documentation/perf-data.txt +7 -0
index 20f178d61ed7..578ba6357183 100644
@@ -30,6 +30,13 @@ OPTIONS for 'convert'
 --to-json::
 	Triggers JSON conversion. Specify the JSON filename to output.
 
+--to-trace-dat::
+	Triggers trace.dat conversion. Converts perf.data tracepoint events
+	to trace.dat format v7, compatible with trace-cmd and KernelShark.
+	Only PERF_TYPE_TRACEPOINT events are converted. Specify the
+	trace.dat filename to output. Requires libtraceevent support.
+	Mutually exclusive with --to-ctf and --to-json.
+
 --tod::
 	Convert time to wall clock time.
 
tools/perf/builtin-data.c +38 -2
index 4c08ccb8c06b..194537503ba9 100644
@@ -30,6 +30,11 @@ static const char *data_usage[] = {
 
 static const char *to_json;
 static const char *to_ctf;
+
+#ifdef HAVE_LIBTRACEEVENT
+static const char *trace_dat_output;
+#endif
+
 static struct perf_data_convert_opts opts = {
 	.force = false,
 	.all = false,
@@ -48,6 +53,10 @@ static const struct option data_options[] = {
 		OPT_BOOLEAN(0, "all", &opts.all, "Convert all events"),
 		OPT_STRING(0, "time", &opts.time_str, "str",
 			   "Time span of interest (start,stop)"),
+#ifdef HAVE_LIBTRACEEVENT
+		OPT_STRING(0, "to-trace-dat", &trace_dat_output,
+			   "file", "Convert to trace.dat format using perf.data tracepoints"),
+#endif
 		OPT_END()
 	};
 
@@ -65,16 +74,43 @@ static int cmd_data_convert(int argc, const char **argv)
 		pr_err("You cannot specify both --to-ctf and --to-json.\n");
 		return -1;
 	}
+#ifdef HAVE_LIBTRACEEVENT
+	if (trace_dat_output && (to_json || to_ctf)) {
+		pr_err("You cannot specify --to-trace-dat with --to-ctf or --to-json.\n");
+		return -1;
+	}
+#endif
+
 #ifdef HAVE_LIBBABELTRACE_SUPPORT
+	#ifdef HAVE_LIBTRACEEVENT
+	if (!to_json && !to_ctf && !trace_dat_output) {
+		pr_err("You must specify one of --to-ctf, --to-json, or --to-trace-dat.\n");
+		return -1;
+	}
+	#else
 	if (!to_json && !to_ctf) {
 		pr_err("You must specify one of --to-ctf or --to-json.\n");
 		return -1;
 	}
+	#endif
 #else
+	#ifdef HAVE_LIBTRACEEVENT
+	if (!to_json && !trace_dat_output) {
+		pr_err("You must specify --to-json or --to-trace-dat.\n");
+		return -1;
+	}
+	#else
 	if (!to_json) {
 		pr_err("You must specify --to-json.\n");
-	return -1;
-}
+		return -1;
+	}
+	#endif
+#endif
+
+#ifdef HAVE_LIBTRACEEVENT
+	if (trace_dat_output)
+		return trace_convert__perf2dat(input_name ? input_name : "perf.data",
+					       trace_dat_output, &opts);
 #endif
 
 	if (to_json)
tools/perf/util/trace-dat.c +3 -0
index f71e03716e27..5598496c9a61 100644
@@ -861,6 +861,9 @@ void trace_dat__free_cpu_buffers(void)
 	for (cpu = 0; cpu < trace_dat_nr_cpus; cpu++) {
 		int i;
 
+		if (!trace_cpu_data[cpu].events)
+			continue;
+
 		for (i = 0; i < trace_cpu_data[cpu].count; i++)
 			free(trace_cpu_data[cpu].events[i].raw);
 		free(trace_cpu_data[cpu].events);
Signature · 1 line
2.47.3
#5 lore raw

RFC PATCH v2 5/5 perf test: Add test validating trace.dat generated by 'perf data convert --to-trace-dat'

In reply to Tanushree Shah

Add a shell test covering perf data convert --to-trace-dat, alongside the existing --to-json and --to-ctf tests.

The test:
- skips if perf isn't linked with libtraceevent
- converts a plain sched:sched_switch recording to trace.dat
- converts a pipe-mode recording (perf record -o - | perf data convert -i -)
- converts a recording with both tracepoint and non-tracepoint events (sched:sched_switch,cycles)
- validates the resulting file with trace-cmd report if trace-cmd is installed, otherwise falls back to a basic non-empty-file check

---
 ...rf_data_converter_tracepoints_trace_dat.sh | 169 ++++++++++++++++++
 1 file changed, 169 insertions(+)
 create mode 100755 tools/perf/tests/shell/test_perf_data_converter_tracepoints_trace_dat.sh
tools/perf/tests/shell/test_perf_data_converter_tracepoints_trace_dat.sh new file +169 -0
new file mode 100755
index 000000000000..9ca6432618cb
@@ -0,0 +1,169 @@
+#!/bin/bash
+# 'perf data convert --to-trace-dat' command test
+# SPDX-License-Identifier: GPL-2.0-or-later
+#
+# Copyright 2026, IBM Corporation
+# Author: Tanushree Shah <[email protected]>
+
+set -e
+
+err=0
+
+perfdata=$(mktemp /tmp/__perf_test.perf.data.XXXXX)
+result=$(mktemp /tmp/__perf_test.output.trace.dat.XXXXX)
+
+cleanup()
+{
+	rm -f "${perfdata}"
+	rm -f "${result}"
+	trap - exit term int
+}
+
+trap_cleanup()
+{
+	echo "Unexpected signal in ${FUNCNAME[1]}"
+	cleanup
+	exit 1
+}
+trap trap_cleanup exit term int
+
+# Check if libtraceevent support is available
+if ! perf check feature libtraceevent
+then
+	echo "perf not linked with libtraceevent, skipping test"
+	exit 2
+fi
+
+# Check if trace-cmd is available for validation
+have_trace_cmd=0
+if command -v trace-cmd >/dev/null 2>&1; then
+	have_trace_cmd=1
+fi
+
+check_sched_switch()
+{
+	if ! perf list | grep -q 'sched:sched_switch'
+	then
+		echo "sched:sched_switch tracepoint not available, skipping test"
+		exit 2
+	fi
+}
+
+test_trace_converter_command()
+{
+	echo "Testing Perf Data Conversion Command to trace.dat"
+
+	if ! perf record -e sched:sched_switch -o "$perfdata" -- sleep 0.1
+	then
+		echo "Failed to record perf data"
+		err=1
+		return
+	fi
+
+	rm -f "$result"
+
+	if ! perf data convert --to-trace-dat "$result" --force -i "$perfdata"
+	then
+		echo "Perf Data Converter Command to trace.dat [FAILED]"
+		err=1
+		return
+	fi
+
+	if [ -f "$result" ] && [ -s "$result" ] ; then
+		echo "Perf Data Converter Command to trace.dat [SUCCESS]"
+	else
+		echo "Perf Data Converter Command to trace.dat [FAILED]"
+		err=1
+	fi
+}
+
+test_trace_converter_pipe()
+{
+	echo "Testing Perf Data Conversion Command to trace.dat (Pipe mode)"
+
+	rm -f "$result"
+
+	if ! perf record -e sched:sched_switch -o - -- sleep 0.1 | \
+	   perf data convert --to-trace-dat "$result" --force -i -
+	then
+		echo "Perf Data Converter Command to trace.dat (Pipe mode) [FAILED]"
+		err=1
+		return
+	fi
+
+	if [ -f "$result" ] && [ -s "$result" ]; then
+		echo "Perf Data Converter Command to trace.dat (Pipe mode) [SUCCESS]"
+	else
+		echo "Perf Data Converter Command to trace.dat (Pipe mode) [FAILED]"
+		err=1
+	fi
+}
+
+test_trace_converter_mixed_events()
+{
+	echo "Testing Perf Data Conversion with tracepoint and non-tracepoint events"
+
+	# Record both tracepoint and non-tracepoint events
+	if ! perf record -e sched:sched_switch,cycles -o "$perfdata" -- sleep 0.1
+	then
+		echo "Failed to record perf data (mixed events)"
+		err=1
+		return
+	fi
+
+	rm -f "$result"
+
+	if ! perf data convert --to-trace-dat "$result" --force -i "$perfdata"
+	then
+		echo "Perf Data Converter with mixed events [FAILED]"
+		err=1
+		return
+	fi
+
+	if [ -f "$result" ] && [ -s "$result" ] ; then
+		echo "Perf Data Converter with mixed events [SUCCESS]"
+	else
+		echo "Perf Data Converter with mixed events [FAILED]"
+		err=1
+	fi
+}
+
+validate_trace_format()
+{
+	echo "Validating Perf Data Converted trace.dat file"
+
+	if [ ! -f "$result" ] ; then
+		echo "File not found [FAILED]"
+		err=1
+		return
+	fi
+
+	# If trace-cmd is available, use it to validate the file
+	if [ "$have_trace_cmd" -eq 1 ] ; then
+		if trace-cmd report -i "$result" >/dev/null 2>&1 ; then
+			echo "trace-cmd can read the file [SUCCESS]"
+		else
+			echo "trace-cmd cannot read the file [FAILED]"
+			err=1
+		fi
+	else
+		# Without trace-cmd, just check file exists and has content
+		if [ -s "$result" ] ; then
+			echo "Output file exists and has content [SUCCESS]"
+		else
+			echo "Output file is empty [FAILED]"
+			err=1
+		fi
+	fi
+}
+
+check_sched_switch
+test_trace_converter_command
+validate_trace_format
+test_trace_converter_pipe
+validate_trace_format
+test_trace_converter_mixed_events
+validate_trace_format
+
+cleanup
+exit ${err}
Signature · 1 line
2.47.3
reviewer
#6 lore raw
In reply to Tanushree Shah
Quoted 231 lines · Tanushree Shah

This RFC patch series introduces support for converting perf.data files containing tracepoint events into trace.dat format, enabling seamless visualization and analysis using KernelShark.

======================
Background and Motivation
======================

Currently, perf and trace-cmd operate as separate tracing ecosystems with incompatible data formats. Users who collect tracepoint data with 'perf record' cannot easily visualize it in KernelShark's graphical timeline view or leverage trace-cmd's analysis capabilities.

This creates workflow friction when users need to:

- Visualize perf tracepoint data in KernelShark's interactive graphical timeline
- Share trace data between perf and trace-cmd workflows and toolchains
- Perform architecture-independent conversion and analysis of traces

This conversion bridge eliminates these barriers by enabling seamless data exchange between perf and trace-cmd ecosystems, allowing users to choose the best tool for each analysis phase.

======================
Implementation Overview
======================

The series implements the trace.dat file format specification (version 7) within perf's data conversion framework.

**Patch 1/5: Core trace.dat Export Infrastructure** Introduces util/trace-dat.c and util/trace-dat.h implementing:
- Per-CPU raw event buffer management (init, collect, free)
- Ftrace ring buffer page construction
- trace.dat section writers (strings, options, flyrecord sections)

**Patch 2/5: Metadata Integration**
Extends util/trace-event-read.c to write trace.dat metadata during perf.data
parsing:
- Initial format header (magic, version, endian, page size, compression)
- Section 16: HEADER INFO (header_page + header_event)
- Section 17: FTRACE EVENT FORMATS
- Section 18: EVENT FORMATS (per system/event format files)
- Section 19: KALLSYMS
- Section 21: CMDLINES
- Section 15: STRINGS (written last after all sections)

**Patch 3/5: Conversion Backend**
Implements util/data-convert-trace.c with trace_convert__perf2dat() function:
- Processes PERF_TYPE_TRACEPOINT samples via process_sample_event()
- Collects raw event data per-CPU using trace_dat__collect_cpu_event()
- Writes OPTIONS sections (CPUCOUNT, TRACECLOCK, metadata offsets)
- Writes FLYRECORD section with per-CPU ring buffer pages

**Patch 4/5: User Interface**
Extends tools/perf/builtin-data.c with --to-trace-dat option:
- Adds command-line option for trace.dat output
- Mutually exclusive with --to-ctf and --to-json
- Calls trace_convert__perf2dat() to perform conversion

**Patch 5/5: Shell Test**
Adds a shell test (tools/perf/tests/shell/) covering normal tracepoint recordings, pipe mode and mixed tracepoint/non-tracepoint recording conversions and --force flag behaviour.

======================
Current Implementation Details
======================

**trace.dat Format Version:**
The implementation currently targets trace.dat format version 7, which is the stable version supported by current trace-cmd releases (v3.x). This version is hardcoded to ensure compatibility with existing trace-cmd and KernelShark installations. Future enhancements could add version negotiation or support for newer format versions as they become standardized.

**Compression Strategy:**
Compression is explicitly disabled (set to NONE) in the generated trace.dat files.
This design choice:
- Simplifies the initial implementation and testing
- Ensures maximum compatibility across trace-cmd versions
- Avoids external compression library dependencies

Future work could add support for various compression algorithms (zlib, zstd, lz4) with runtime selection via command-line options, significantly reducing file sizes for large traces.

======================
Usage Example
======================

```bash
*Record tracepoint events with perf*
perf record -e sched:sched_switch -e sched:sched_wakeup -a sleep 10

*Convert to trace.dat format*
perf data convert --to-trace-dat=output.dat

*Verify trace.dat structure*
trace-cmd dump --summary output.dat

*Analyze with trace-cmd*
trace-cmd report output.dat

*Visualize in KernelShark*
kernelshark output.dat
```

**Conversion Output:**

```
[ perf data convert: Converted 'perf.data' into trace.dat format
'output.dat' ]
[ perf data convert: Converted 2684 events ]
```
**trace-cmd dump --summary Output:**
```
 Tracing meta data in file output.dat:
        [Initial format]
                7       [Version]
                0       [Little endian]
                8       [Bytes in a long]
                65536   [Page size, bytes]
                none    [Compression algorithm]
                        [Compression version]
        [buffer "", "local" clock, 65536 page size, 16 cpus, 1048576 bytes
    flyrecord data]
        [10 options]
        [Saved command lines, 0 bytes]
        [Kallsyms, 0 bytes]
        [Ftrace format, 0 events]
        [Header page, 206 bytes]
        [Header event, 205 bytes]
        [Events format, 1 systems]
        [9 sections]
```

======================
Testing and Verification
======================

The series has been extensively tested with:
- Various tracepoint events (sched, irq, syscalls, block I/O)
- Mixed recordings containing both tracepoint and non-tracepoint events only tracepoints converted)
- Verification with trace-cmd report and KernelShark visualization
- Memory leak testing with Valgrind (0 bytes leaked).
- Cross-architecture testing: v1 tested x86_64 and ppc64le. v2 adds s390 (big-endian) perf.data converted on both ppc64le (little-endian) and x86_64 (little-endian) hosts, in addition to same-arch x86_64 (LE->LE) and ppc64le (BE->BE) conversion.
- Pipe mode support has been tested end-to-end. (in v2)

All generated trace.dat files successfully open in:
- trace-cmd report (v3.1+)
- KernelShark (v2.0+)

======================
Next Steps
======================

We would highly appreciate reviews, comments, and feedback on:
- The overall architectural approach and integration points
- Compatibility considerations with trace-cmd ecosystem
- Performance characteristics for large-scale traces
- Additional use cases or workflow scenarios
- Future enhancement priorities

---
Changes in v2

Addressing the Sashiko AI review findings on v1:

Cross-arch correctness:
- Introduce to_file_u16/u32/u64 helpers (wrapping tep_read_number()) to write all multi-byte fields in the recorded machine's byte order; apply throughout metadata sections and flyrecord page/record headers (ts, commit, TIME_EXTEND, large-event data_len).
- Fix flyrecord record header bit layout for big-endian files. The record header word bit layout differs by file endianness, matching kbuffer-parse.c type_len4host()/ts4host():
LE: type_len in bits [4:0], time_delta in bits [31:5]
BE: type_len in bits [31:27], time_delta in bits [26:0]

Pipe mode:
- Add process_attr(), process_feature(), and process_tracing_data() callbacks required for pipe mode operation.
- Defer CPU buffer initialisation until the first tracepoint sample, after process_feature()/process_tracing_data() have populated the session header. This ensures the recorded machine's CPU count is used rather than the host's - critical for cross-platform analysis.

Format compliance:
- Implement TIME_EXTEND records for timestamp deltas >27 bits to prevent silent truncation and maintain chronological ordering.
- Fix large event encoding (>=29 words): use type_len=0 with a separate 32-bit length word, avoiding collision with reserved types (PADDING=29, TIME_EXTEND=30, TIME_STAMP=31).
- Add bounds check rejecting records larger than a page payload before batching, preventing heap overflow in trace_dat__write_page().
- Fix flyrecord section_size to exclude the 16-byte section header, matching trace.dat specification and trace-cmd behaviour.

CLI behavior:
- Fix --force flag: open with O_CREAT|O_EXCL when force is not set, failing with -EEXIST instead of silently overwriting existing files.

Memory safety:
- Fix realloc overwrite of cpu_events->events and page_records on
failure: use temporary pointers, only commit on success.
- Fix use-after-free/double-free in sequential page_records realloc
failure: replace with malloc+memcpy+free pattern.
- Fix section_size computed from before section header position.
- Add NULL checks for get_tracing_file(), calloc() padding, and trace_dat_options_offset assignment on write failure.
- Use goto out_free on record allocation failure to avoid leaking accumulated page_records entries.
- Replace direct read() with do_read() in read_proc_kallsyms() to handle short reads correctly.
- On fwrite failure, set trace_dat_write_failed and continue parsing so that perf.data processing completes normally.

Testing (new in v2):
- Add shell test covering conversion, trace-cmd dump validation, sched_switch event verification, and --force flag behaviour.

Quoted 5 lines

Documentation (new in v2):
- Add documentation for 'perf data convert --to-trace-dat', covering usage and supported options.

v1: https://lore.kernel.org/linux-perf-users/[email protected]/

Thanks again Tanushree for pursuing this. I see on sashiko.dev that sashiko couldn't review the v2 patches because they failed to apply:
https://sashiko.dev/#/log/baseline/53604/2
The SHA of perf-tools-next that sashiko.dev was trying was d04ef71492fad7230d474efe33d05f4c0563d409, which corresponds to "perf
machine: Check snprintf truncation for guest kallsyms path," added 4 days ago:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/commit/?h=perf-tools-next&id=d04ef71492fad7230d474efe33d05f4c0563d409
Could you rebase the patches so that we can benefit from the Sashiko review?

Thanks,
Ian

Quoted 28 lines

Tanushree Shah (5):
perf/trace-dat: Add trace.dat export infrastructure perf/trace-event: Write trace.dat metadata sections during parsing perf data-convert: Add perf.data to trace.dat conversion backend perf data: Add --to-trace-dat option for converting perf.data
tracepoint events into trace.dat format
perf test: Add test validating trace.dat generated by 'perf data

    convert --to-trace-dat'
 tools/perf/Documentation/perf-data.txt        |   7 +
 tools/perf/builtin-data.c                     |  40 +-
 ...rf_data_converter_tracepoints_trace_dat.sh | 169 ++++
 tools/perf/util/Build                         |   2 +
 tools/perf/util/data-convert-trace.c          | 240 +++++
 tools/perf/util/data-convert.h                |   4 +
 tools/perf/util/trace-dat.c                   | 876 ++++++++++++++++++
 tools/perf/util/trace-dat.h                   | 113 +++
 tools/perf/util/trace-event-read.c            | 307 +++++-
 9 files changed, 1749 insertions(+), 9 deletions(-)
 create mode 100755 tools/perf/tests/shell/test_perf_data_converter_tracepoints_trace_dat.sh
 create mode 100644 tools/perf/util/data-convert-trace.c
 create mode 100644 tools/perf/util/trace-dat.c
 create mode 100644 tools/perf/util/trace-dat.h

--
2.47.3

reviewer
#7 lore raw
Quoted 1 line · Ian Rogers

On Sun, Aug 2, 2026 at 11:08 AM Tanushree Shah <[email protected]> wrote:

...

Quoted 11 lines
Quoted 1 line

v1: https://lore.kernel.org/linux-perf-users/[email protected]/

Thanks again Tanushree for pursuing this. I see on sashiko.dev that sashiko couldn't review the v2 patches because they failed to apply:
https://sashiko.dev/#/log/baseline/53604/2
The SHA of perf-tools-next that sashiko.dev was trying was d04ef71492fad7230d474efe33d05f4c0563d409, which corresponds to "perf
machine: Check snprintf truncation for guest kallsyms path," added 4 days ago:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/commit/?h=perf-tools-next&id=d04ef71492fad7230d474efe33d05f4c0563d409
Could you rebase the patches so that we can benefit from the Sashiko review?

Fwiw, I went through the individual patches in case I could give some useful early feedback. Everything looks great, and thanks for adding the test!

Thanks,
Ian

#8 lore raw

Thank you Ian for the review, glad it looks good!

Thanks,
Tanushree Shah

Quoted 21 lines · Ian Rogers
Quoted 1 line · Ian Rogers

On Sun, Aug 2, 2026 at 11:08 AM Tanushree Shah <[email protected]> wrote:

...

Quoted 11 lines
Quoted 1 line

v1: https://lore.kernel.org/linux-perf-users/[email protected]/

Thanks again Tanushree for pursuing this. I see on sashiko.dev that sashiko couldn't review the v2 patches because they failed to apply:
https://sashiko.dev/#/log/baseline/53604/2
The SHA of perf-tools-next that sashiko.dev was trying was d04ef71492fad7230d474efe33d05f4c0563d409, which corresponds to "perf
machine: Check snprintf truncation for guest kallsyms path," added 4 days ago:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/commit/?h=perf-tools-next&id=d04ef71492fad7230d474efe33d05f4c0563d409
Could you rebase the patches so that we can benefit from the Sashiko review?

Fwiw, I went through the individual patches in case I could give some useful early feedback. Everything looks great, and thanks for adding the test!

Thanks,
Ian

#9 lore raw
In reply to Ian Rogers

Hello Ian, Thank you very much for the feedback.

yes, v2 needed a rebase. I had already sent V3 patches on Monday. Below is the link:
https://lore.kernel.org/all/[email protected]/.

Thanks.
Tanushree Shah

Quoted 279 lines · Ian Rogers
Quoted 231 lines · Tanushree Shah

This RFC patch series introduces support for converting perf.data files containing tracepoint events into trace.dat format, enabling seamless visualization and analysis using KernelShark.

======================
Background and Motivation
======================

Currently, perf and trace-cmd operate as separate tracing ecosystems with incompatible data formats. Users who collect tracepoint data with 'perf record' cannot easily visualize it in KernelShark's graphical timeline view or leverage trace-cmd's analysis capabilities.

This creates workflow friction when users need to:

- Visualize perf tracepoint data in KernelShark's interactive graphical timeline
- Share trace data between perf and trace-cmd workflows and toolchains
- Perform architecture-independent conversion and analysis of traces

This conversion bridge eliminates these barriers by enabling seamless data exchange between perf and trace-cmd ecosystems, allowing users to choose the best tool for each analysis phase.

======================
Implementation Overview
======================

The series implements the trace.dat file format specification (version 7) within perf's data conversion framework.

**Patch 1/5: Core trace.dat Export Infrastructure** Introduces util/trace-dat.c and util/trace-dat.h implementing:
- Per-CPU raw event buffer management (init, collect, free)
- Ftrace ring buffer page construction
- trace.dat section writers (strings, options, flyrecord sections)

**Patch 2/5: Metadata Integration**
Extends util/trace-event-read.c to write trace.dat metadata during perf.data
parsing:
- Initial format header (magic, version, endian, page size, compression)
- Section 16: HEADER INFO (header_page + header_event)
- Section 17: FTRACE EVENT FORMATS
- Section 18: EVENT FORMATS (per system/event format files)
- Section 19: KALLSYMS
- Section 21: CMDLINES
- Section 15: STRINGS (written last after all sections)

**Patch 3/5: Conversion Backend**
Implements util/data-convert-trace.c with trace_convert__perf2dat() function:
- Processes PERF_TYPE_TRACEPOINT samples via process_sample_event()
- Collects raw event data per-CPU using trace_dat__collect_cpu_event()
- Writes OPTIONS sections (CPUCOUNT, TRACECLOCK, metadata offsets)
- Writes FLYRECORD section with per-CPU ring buffer pages

**Patch 4/5: User Interface**
Extends tools/perf/builtin-data.c with --to-trace-dat option:
- Adds command-line option for trace.dat output
- Mutually exclusive with --to-ctf and --to-json
- Calls trace_convert__perf2dat() to perform conversion

**Patch 5/5: Shell Test**
Adds a shell test (tools/perf/tests/shell/) covering normal tracepoint recordings, pipe mode and mixed tracepoint/non-tracepoint recording conversions and --force flag behaviour.

======================
Current Implementation Details
======================

**trace.dat Format Version:**
The implementation currently targets trace.dat format version 7, which is the stable version supported by current trace-cmd releases (v3.x). This version is hardcoded to ensure compatibility with existing trace-cmd and KernelShark installations. Future enhancements could add version negotiation or support for newer format versions as they become standardized.

**Compression Strategy:**
Compression is explicitly disabled (set to NONE) in the generated trace.dat files.
This design choice:
- Simplifies the initial implementation and testing
- Ensures maximum compatibility across trace-cmd versions
- Avoids external compression library dependencies

Future work could add support for various compression algorithms (zlib, zstd, lz4) with runtime selection via command-line options, significantly reducing file sizes for large traces.

======================
Usage Example
======================

```bash
*Record tracepoint events with perf*
perf record -e sched:sched_switch -e sched:sched_wakeup -a sleep 10

*Convert to trace.dat format*
perf data convert --to-trace-dat=output.dat

*Verify trace.dat structure*
trace-cmd dump --summary output.dat

*Analyze with trace-cmd*
trace-cmd report output.dat

*Visualize in KernelShark*
kernelshark output.dat
```

**Conversion Output:**

```
[ perf data convert: Converted 'perf.data' into trace.dat format
'output.dat' ]
[ perf data convert: Converted 2684 events ]
```
**trace-cmd dump --summary Output:**
```
  Tracing meta data in file output.dat:
         [Initial format]
                 7       [Version]
                 0       [Little endian]
                 8       [Bytes in a long]
                 65536   [Page size, bytes]
                 none    [Compression algorithm]
                         [Compression version]
         [buffer "", "local" clock, 65536 page size, 16 cpus, 1048576 bytes
     flyrecord data]
         [10 options]
         [Saved command lines, 0 bytes]
         [Kallsyms, 0 bytes]
         [Ftrace format, 0 events]
         [Header page, 206 bytes]
         [Header event, 205 bytes]
         [Events format, 1 systems]
         [9 sections]
```

======================
Testing and Verification
======================

The series has been extensively tested with:
- Various tracepoint events (sched, irq, syscalls, block I/O)
- Mixed recordings containing both tracepoint and non-tracepoint events only tracepoints converted)
- Verification with trace-cmd report and KernelShark visualization
- Memory leak testing with Valgrind (0 bytes leaked).
- Cross-architecture testing: v1 tested x86_64 and ppc64le. v2 adds s390 (big-endian) perf.data converted on both ppc64le (little-endian) and x86_64 (little-endian) hosts, in addition to same-arch x86_64 (LE->LE) and ppc64le (BE->BE) conversion.
- Pipe mode support has been tested end-to-end. (in v2)

All generated trace.dat files successfully open in:
- trace-cmd report (v3.1+)
- KernelShark (v2.0+)

======================
Next Steps
======================

We would highly appreciate reviews, comments, and feedback on:
- The overall architectural approach and integration points
- Compatibility considerations with trace-cmd ecosystem
- Performance characteristics for large-scale traces
- Additional use cases or workflow scenarios
- Future enhancement priorities

---
Changes in v2

Addressing the Sashiko AI review findings on v1:

Cross-arch correctness:
- Introduce to_file_u16/u32/u64 helpers (wrapping tep_read_number()) to write all multi-byte fields in the recorded machine's byte order; apply throughout metadata sections and flyrecord page/record headers (ts, commit, TIME_EXTEND, large-event data_len).
- Fix flyrecord record header bit layout for big-endian files. The record header word bit layout differs by file endianness, matching kbuffer-parse.c type_len4host()/ts4host():
LE: type_len in bits [4:0], time_delta in bits [31:5]
BE: type_len in bits [31:27], time_delta in bits [26:0]

Pipe mode:
- Add process_attr(), process_feature(), and process_tracing_data() callbacks required for pipe mode operation.
- Defer CPU buffer initialisation until the first tracepoint sample, after process_feature()/process_tracing_data() have populated the session header. This ensures the recorded machine's CPU count is used rather than the host's - critical for cross-platform analysis.

Format compliance:
- Implement TIME_EXTEND records for timestamp deltas >27 bits to prevent silent truncation and maintain chronological ordering.
- Fix large event encoding (>=29 words): use type_len=0 with a separate 32-bit length word, avoiding collision with reserved types (PADDING=29, TIME_EXTEND=30, TIME_STAMP=31).
- Add bounds check rejecting records larger than a page payload before batching, preventing heap overflow in trace_dat__write_page().
- Fix flyrecord section_size to exclude the 16-byte section header, matching trace.dat specification and trace-cmd behaviour.

CLI behavior:
- Fix --force flag: open with O_CREAT|O_EXCL when force is not set, failing with -EEXIST instead of silently overwriting existing files.

Memory safety:
- Fix realloc overwrite of cpu_events->events and page_records on
failure: use temporary pointers, only commit on success.
- Fix use-after-free/double-free in sequential page_records realloc
failure: replace with malloc+memcpy+free pattern.
- Fix section_size computed from before section header position.
- Add NULL checks for get_tracing_file(), calloc() padding, and trace_dat_options_offset assignment on write failure.
- Use goto out_free on record allocation failure to avoid leaking accumulated page_records entries.
- Replace direct read() with do_read() in read_proc_kallsyms() to handle short reads correctly.
- On fwrite failure, set trace_dat_write_failed and continue parsing so that perf.data processing completes normally.

Testing (new in v2):
- Add shell test covering conversion, trace-cmd dump validation, sched_switch event verification, and --force flag behaviour.

Quoted 5 lines

Documentation (new in v2):
- Add documentation for 'perf data convert --to-trace-dat', covering usage and supported options.

v1: https://lore.kernel.org/linux-perf-users/[email protected]/

Thanks again Tanushree for pursuing this. I see on sashiko.dev that sashiko couldn't review the v2 patches because they failed to apply:
https://sashiko.dev/#/log/baseline/53604/2
The SHA of perf-tools-next that sashiko.dev was trying was d04ef71492fad7230d474efe33d05f4c0563d409, which corresponds to "perf
machine: Check snprintf truncation for guest kallsyms path," added 4 days ago:
https://web.git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/commit/?h=perf-tools-next&id=d04ef71492fad7230d474efe33d05f4c0563d409
Could you rebase the patches so that we can benefit from the Sashiko review?

Thanks,
Ian

Quoted 28 lines

Tanushree Shah (5):
perf/trace-dat: Add trace.dat export infrastructure perf/trace-event: Write trace.dat metadata sections during parsing perf data-convert: Add perf.data to trace.dat conversion backend perf data: Add --to-trace-dat option for converting perf.data
tracepoint events into trace.dat format
perf test: Add test validating trace.dat generated by 'perf data

     convert --to-trace-dat'
  tools/perf/Documentation/perf-data.txt        |   7 +
  tools/perf/builtin-data.c                     |  40 +-
  ...rf_data_converter_tracepoints_trace_dat.sh | 169 ++++
  tools/perf/util/Build                         |   2 +
  tools/perf/util/data-convert-trace.c          | 240 +++++
  tools/perf/util/data-convert.h                |   4 +
  tools/perf/util/trace-dat.c                   | 876 ++++++++++++++++++
  tools/perf/util/trace-dat.h                   | 113 +++
  tools/perf/util/trace-event-read.c            | 307 +++++-
  9 files changed, 1749 insertions(+), 9 deletions(-)

create mode 100755 tools/perf/tests/shell/test_perf_data_converter_tracepoints_trace_dat.sh create mode 100644 tools/perf/util/data-convert-trace.c
create mode 100644 tools/perf/util/trace-dat.c
create mode 100644 tools/perf/util/trace-dat.h

--
2.47.3