From: Shani Peretz <[email protected]> Add a Python script that parses the history dump of mbufs generated by rte_mbuf_history_dump() and related functions, and presents it in a human-readable format.
If an operation ID is repeated, such as in the case of a double free, it will be highlighted in red and listed at the end of the file. Signed-off-by: Shani Peretz <[email protected]> --- doc/guides/howto/debug_troubleshoot.rst | 3 +- doc/guides/prog_guide/mbuf_lib.rst | 2 + doc/guides/rel_notes/release_25_11.rst | 1 + lib/mbuf/rte_mbuf_history.h | 3 + usertools/dpdk-mbuf-history-parser.py | 166 ++++++++++++++++++++++++ 5 files changed, 174 insertions(+), 1 deletion(-) create mode 100755 usertools/dpdk-mbuf-history-parser.py diff --git a/doc/guides/howto/debug_troubleshoot.rst b/doc/guides/howto/debug_troubleshoot.rst index 16feeb1e54..c19aada6f8 100644 --- a/doc/guides/howto/debug_troubleshoot.rst +++ b/doc/guides/howto/debug_troubleshoot.rst @@ -221,7 +221,8 @@ Memory objects close to NUMA :numref:`dtg_mempool`. by defining the compilation flag ``RTE_MBUF_HISTORY_DEBUG``. Then the libraries will mark the mbufs, and more marks can be added in the application. - Some dump functions must be used to collect the history. + Some dump functions must be used to collect the history, + and a script can parse it. #. Lower performance between the pipeline processing stages can be diff --git a/doc/guides/prog_guide/mbuf_lib.rst b/doc/guides/prog_guide/mbuf_lib.rst index ae72eecc61..382bfbdca4 100644 --- a/doc/guides/prog_guide/mbuf_lib.rst +++ b/doc/guides/prog_guide/mbuf_lib.rst @@ -279,6 +279,8 @@ More marks can be added by the application by calling functions like ``rte_mbuf_history_mark_bulk()``. Then the history can be stored in a file by calling functions like ``rte_mbuf_history_dump_all()``. +The dump file will be easier to read after being processed +by the script ``dpdk-mbuf-history-parser.py``. Use Cases diff --git a/doc/guides/rel_notes/release_25_11.rst b/doc/guides/rel_notes/release_25_11.rst index 6854090096..6c26d3be10 100644 --- a/doc/guides/rel_notes/release_25_11.rst +++ b/doc/guides/rel_notes/release_25_11.rst @@ -65,6 +65,7 @@ New Features Added history dynamic field in mbuf (disabled by default) to store successive states of the mbuf lifecycle. Some functions were added to dump statistics. + A script was added to parse mbuf tracking stored in a file. * **Updated NXP DPAA2 ethernet driver.** diff --git a/lib/mbuf/rte_mbuf_history.h b/lib/mbuf/rte_mbuf_history.h index f5f95a0c60..c4ca38e746 100644 --- a/lib/mbuf/rte_mbuf_history.h +++ b/lib/mbuf/rte_mbuf_history.h @@ -19,6 +19,9 @@ * The history is stored as an atomic value (64-bit) in a dynamic field of the mbuf, * with each event encoded in 4 bits, allowing up to 16 events to be tracked. * Atomic operations ensure thread safety for cloned mbufs accessed by multiple lcores. + * + * After dumping the history in a file, + * the script dpdk-mbuf-history-parser.py can be used for parsing. */ #include <rte_common.h> diff --git a/usertools/dpdk-mbuf-history-parser.py b/usertools/dpdk-mbuf-history-parser.py new file mode 100755 index 0000000000..6d01d0ff67 --- /dev/null +++ b/usertools/dpdk-mbuf-history-parser.py @@ -0,0 +1,166 @@ +#!/usr/bin/env python3 +# SPDX-License-Identifier: BSD-3-Clause +# Copyright (c) 2023 NVIDIA Corporation & Affiliates + +""" +Parse the mbuf history dump generated by rte_mbuf_history_dump() +and related functions, and present it in a human-readable format. +""" + +import argparse +import enum +import os +import re +import sys + +RED = "\033[91m" +RESET = "\033[0m" +ENUM_PATTERN = r"enum\s+rte_mbuf_history_op\s*{([^}]+)}" +VALUE_PATTERN = r"([A-Z_]+)\s*=\s*(\d+),\s*(?:/\*\s*(.*?)\s*\*/)?" +HEADER_FILE = os.path.join( + os.path.dirname(os.path.dirname(__file__)), "lib/mbuf/rte_mbuf_history.h" +) + + +def print_history_sequence(address: str, sequence: list[str]): + max_op_width = max(len(re.sub(r"\x1b\[[0-9;]*m", "", op)) for op in sequence) + op_width = max_op_width + for i in range(0, len(sequence), 4): + chunk = sequence[i : i + 4] + formatted_ops = [f"{op:<{op_width}}" for op in chunk] + line = "" + for j, op in enumerate(formatted_ops): + line += op + if j < len(formatted_ops) - 1: + line += " -> " + if i + 4 < len(sequence): + line += " ->" + print(f"mbuf {address}: " + line) + print() + + +def match_field(match: re.Match) -> tuple[int, str]: + name, value, _ = match.groups() + return (int(value), name.replace("RTE_MBUF_", "")) + + +class HistoryEnum: + def __init__(self, ops: enum.Enum): + self.ops = ops + + @staticmethod + def from_header(header_file: str) -> "HistoryEnum": + with open(header_file, "r") as f: + content = f.read() + + # Extract each enum value and its comment + enum_content = re.search(ENUM_PATTERN, content, re.DOTALL).group(1) + fields = map(match_field, re.finditer(VALUE_PATTERN, enum_content)) + fields = dict({v: k for k, v in fields}) + return HistoryEnum(enum.Enum("HistoryOps", fields)) + + +class HistoryLine: + def __init__(self, address: str, ops: list): + self.address = address + self.ops = ops + + def repeats(self) -> [list[str], str | None]: + repeated = None + sequence = [] + for idx, op in enumerate(self.ops): + if idx > 0 and op == self.ops[idx - 1] and op.name != "NEVER": + sequence[-1] = f"{RED}{op.name}{RESET}" + sequence.append(f"{RED}{op.name}{RESET}") + repeated = op.name + else: + sequence.append(op.name) + return sequence, repeated + + +class HistoryMetrics: + def __init__(self, metrics: dict[str, int]): + self.metrics = metrics + + def max_name_width(self) -> int: + return max(len(name) for name in self.metrics.keys()) + + +class HistoryParser: + def __init__(self): + self.history_enum = HistoryEnum.from_header(HEADER_FILE) + + def parse(self, dump_file: str) -> tuple[list[HistoryLine], "HistoryMetrics"]: + with open(dump_file, "r") as f: + lines = [line for line in f.readlines() if line.strip()] + populated = next(line for line in lines if " populated =" in line) + metrics_start = lines.index(populated) + + history_lines = lines[3:metrics_start] + metrics_lines = lines[metrics_start:] + return (self._parse_history(history_lines), self._parse_metrics(metrics_lines)) + + def _parse_metrics(self, lines: list[str]) -> HistoryMetrics: + metrics = {} + for line in lines: + key, value = line.split("=", 1) + metrics[key] = int(value) + return HistoryMetrics(metrics) + + def _parse_history(self, lines: list[str]) -> list[HistoryLine]: + # Parse the format "mbuf 0x1054b9980: 0000000000000065" + history_lines = [] + for line in lines: + address = line.split(":")[0].split("mbuf ")[1] + history = line.split(":")[1] + history_lines.append( + HistoryLine(address=address, ops=self._parse(int(history, 16))) + ) + return history_lines + + def _parse(self, history: int) -> list[str]: + ops = [] + for _ in range(16): # 64 bits / 4 bits = 16 possible operations + op = history & 0xF # Extract lowest 4 bits + if op == 0: + break + ops.append(self.history_enum.ops(op)) + history >>= 4 + + ops.reverse() + return ops + + +def print_history_lines(history_lines: list[HistoryLine]): + lines = [(line.address, line.repeats()) for line in history_lines] + + for address, (sequence, _) in lines: + print_history_sequence(address, sequence) + + print("=== Violations ===") + for address, (sequence, repeated) in lines: + if repeated: + print(f"mbuf {address} has repeated ops: {RED}{repeated}{RESET}") + + +def print_metrics(metrics: HistoryMetrics): + print("=== Metrics Summary ===") + for name, value in metrics.metrics.items(): + print(f"{name + '':<{metrics.max_name_width() + 2}} {value}") + + +def main(): + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument('history_file') + args = parser.parse_args() + + history_parser = HistoryParser() + history_lines, metrics = history_parser.parse(args.history_file) + + print_history_lines(history_lines) + print() + print_metrics(metrics) + + +if __name__ == "__main__": + main() -- 2.51.0
