perf-prof 标签 - Gitee.com

v1.9.0

perf-prof v1.9.0

DWARF unwind support:
- Add DWARF-based user-space stack unwinding via libunwind
- Integrate --user-callchain=dwarf[,size] option for all supported profilers:
  profile, trace, split-lock, task-state, num-dist, multi-trace,
  page-faults, hrtimer, breakpoint, kmemleak, kmemprof, flame_graph
- Support DWARF unwind in Python PerfEvent callchain and print
- Enable per-event DWARF sampling in tp_evsel_new
- Return zero for unmapped memory reads in access_mem to handle
  DW_EH_PE_indirect personality routine references

Profiler changes:
- Remove sched-migrate profiler

Bug fixes:
- Fix perf_event_member_offset for variable-size members
- Fix unused variable on aarch64 in dwarf_unwind
- Use prof_dev_env2attr in hrtimer to enable DWARF-required sampling
- Handle nested function definitions in symbol lookup
- Use buildid_obj for unwind data lookup

Packaging:
- Add libunwind as required dependency for RPM builds
- Support specifying architecture in download-python-standalone.sh

Documentation:
- Add --user-callchain=dwarf option to all profiler documentation
- Update docs/ and skills/ references for DWARF support

Tests:
- Add DWARF unwind callchain test cases for all supported profilers

1e02d8e

2026-03-18 10:25

下载

v1.8.2

c8f1006

2026-03-04 23:56

下载

v1.8.1

14d6885

2026-03-02 14:46

下载

v1.8.0

perf-prof v1.8.0

Python scripting profiler:
- Add Python scripting profiler for custom event processing
- PerfEvent type with lazy field evaluation and callchain support
- Built-in perf_prof module with print_event function
- Profiler event source (dev_tp) support
- Script template generation, alias attribute, minevtime support
- Python 3.7~3.15 version compatibility

Event processing improvements:
- Add print_event callback for unified event printing
- Move event preprocessing into perf_event_forward()
- Add external field for fast evsel -> tp/dev access
- Simplify tp lookup using tp_from_evsel() across profilers

kmemleak enhancements:
- Add --comm option for allocator process names in report
- Per-comm leaked bytes tracking
- Replace rblist with rb_root/rb_root_cached
- Remove gc_free and enable order by default
- Defer bytes_alloc computation from sample to collect_leaked_bytes

Build system:
- Support python-build-standalone with PYTHON= option
- Add NO_LIBPYTHON switch and fix CFLAGS_REMOVE_ semantics
- Move traceevent plugins under /usr/lib64/perf-prof/ directory

Library updates:
- Update libtraceevent to 1.9.0
- Merge upstream libperf patches for evlist, evsel, and mmap

Other:
- Add single-entry cache to global_comm_get()
- Order: allocate event_copy on demand for out-of-order fix
- Add perf_event_member_cache and member access helpers

781bd6e

2026-02-28 18:54

下载

v1.7.2

78f68d5

2026-01-09 17:30

下载

v1.7.1

8ec3ffb

2026-01-08 11:19

下载

v1.7.0