| 1 | #!/usr/bin/env python3 |
| 2 | |
| 3 | # Print the top N most executed functions in QEMU using perf. |
| 4 | # Syntax: |
| 5 | # topN_perf.py [-h] [-n] <number of displayed top functions> -- \ |
| 6 | # <qemu executable> [<qemu executable options>] \ |
| 7 | # <target executable> [<target executable options>] |
| 8 | # |
| 9 | # [-h] - Print the script arguments help message. |
| 10 | # [-n] - Specify the number of top functions to print. |
| 11 | # - If this flag is not specified, the tool defaults to 25. |
| 12 | # |
| 13 | # Example of usage: |
| 14 | # topN_perf.py -n 20 -- qemu-arm coulomb_double-arm |
| 15 | # |
| 16 | # This file is a part of the project "TCG Continuous Benchmarking". |
| 17 | # |
| 18 | # Copyright (C) 2020 Ahmed Karaman <ahmedkhaledkaraman@gmail.com> |
| 19 | # Copyright (C) 2020 Aleksandar Markovic <aleksandar.qemu.devel@gmail.com> |
| 20 | # |
| 21 | # This program is free software: you can redistribute it and/or modify |
| 22 | # it under the terms of the GNU General Public License as published by |
| 23 | # the Free Software Foundation, either version 2 of the License, or |
| 24 | # (at your option) any later version. |
| 25 | # |
| 26 | # This program is distributed in the hope that it will be useful, |
| 27 | # but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 28 | # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 29 | # GNU General Public License for more details. |
| 30 | # |
| 31 | # You should have received a copy of the GNU General Public License |
| 32 | # along with this program. If not, see <https://www.gnu.org/licenses/>. |
| 33 | |
| 34 | import argparse |
| 35 | import os |
| 36 | import subprocess |
| 37 | import sys |
| 38 | |
| 39 | |
| 40 | # Parse the command line arguments |
| 41 | parser = argparse.ArgumentParser( |
| 42 | usage='topN_perf.py [-h] [-n] <number of displayed top functions > -- ' |
| 43 | '<qemu executable> [<qemu executable options>] ' |
| 44 | '<target executable> [<target executable options>]') |
| 45 | |
| 46 | parser.add_argument('-n', dest='top', type=int, default=25, |
| 47 | help='Specify the number of top functions to print.') |
| 48 | |
| 49 | parser.add_argument('command', type=str, nargs='+', help=argparse.SUPPRESS) |
| 50 | |
| 51 | args = parser.parse_args() |
| 52 | |
| 53 | # Extract the needed variables from the args |
| 54 | command = args.command |
| 55 | top = args.top |
| 56 | |
| 57 | # Insure that perf is installed |
| 58 | check_perf_presence = subprocess.run(["which", "perf"], |
| 59 | stdout=subprocess.DEVNULL) |
| 60 | if check_perf_presence.returncode: |
| 61 | sys.exit("Please install perf before running the script!") |
| 62 | |
| 63 | # Insure user has previllage to run perf |
| 64 | check_perf_executability = subprocess.run(["perf", "stat", "ls", "/"], |
| 65 | stdout=subprocess.DEVNULL, |
| 66 | stderr=subprocess.DEVNULL) |
| 67 | if check_perf_executability.returncode: |
| 68 | sys.exit( |
| 69 | """ |
| 70 | Error: |
| 71 | You may not have permission to collect stats. |
| 72 | |
| 73 | Consider tweaking /proc/sys/kernel/perf_event_paranoid, |
| 74 | which controls use of the performance events system by |
| 75 | unprivileged users (without CAP_SYS_ADMIN). |
| 76 | |
| 77 | -1: Allow use of (almost) all events by all users |
| 78 | Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK |
| 79 | 0: Disallow ftrace function tracepoint by users without CAP_SYS_ADMIN |
| 80 | Disallow raw tracepoint access by users without CAP_SYS_ADMIN |
| 81 | 1: Disallow CPU event access by users without CAP_SYS_ADMIN |
| 82 | 2: Disallow kernel profiling by users without CAP_SYS_ADMIN |
| 83 | |
| 84 | To make this setting permanent, edit /etc/sysctl.conf too, e.g.: |
| 85 | kernel.perf_event_paranoid = -1 |
| 86 | |
| 87 | * Alternatively, you can run this script under sudo privileges. |
| 88 | """ |
| 89 | ) |
| 90 | |
| 91 | # Run perf record |
| 92 | perf_record = subprocess.run((["perf", "record", "--output=/tmp/perf.data"] + |
| 93 | command), |
| 94 | stdout=subprocess.DEVNULL, |
| 95 | stderr=subprocess.PIPE) |
| 96 | if perf_record.returncode: |
| 97 | os.unlink('/tmp/perf.data') |
| 98 | sys.exit(perf_record.stderr.decode("utf-8")) |
| 99 | |
| 100 | # Save perf report output to /tmp/perf_report.out |
| 101 | with open("/tmp/perf_report.out", "w") as output: |
| 102 | perf_report = subprocess.run( |
| 103 | ["perf", "report", "--input=/tmp/perf.data", "--stdio"], |
| 104 | stdout=output, |
| 105 | stderr=subprocess.PIPE) |
| 106 | if perf_report.returncode: |
| 107 | os.unlink('/tmp/perf.data') |
| 108 | output.close() |
| 109 | os.unlink('/tmp/perf_report.out') |
| 110 | sys.exit(perf_report.stderr.decode("utf-8")) |
| 111 | |
| 112 | # Read the reported data to functions[] |
| 113 | functions = [] |
| 114 | with open("/tmp/perf_report.out", "r") as data: |
| 115 | # Only read lines that are not comments (comments start with #) |
| 116 | # Only read lines that are not empty |
| 117 | functions = [line for line in data.readlines() if line and line[0] |
| 118 | != '#' and line[0] != "\n"] |
| 119 | |
| 120 | # Limit the number of top functions to "top" |
| 121 | number_of_top_functions = top if len(functions) > top else len(functions) |
| 122 | |
| 123 | # Store the data of the top functions in top_functions[] |
| 124 | top_functions = functions[:number_of_top_functions] |
| 125 | |
| 126 | # Print table header |
| 127 | print('{:>4} {:>10} {:<30} {}\n{} {} {} {}'.format('No.', |
| 128 | 'Percentage', |
| 129 | 'Name', |
| 130 | 'Invoked by', |
| 131 | '-' * 4, |
| 132 | '-' * 10, |
| 133 | '-' * 30, |
| 134 | '-' * 25)) |
| 135 | |
| 136 | # Print top N functions |
| 137 | for (index, function) in enumerate(top_functions, start=1): |
| 138 | function_data = function.split() |
| 139 | function_percentage = function_data[0] |
| 140 | function_name = function_data[-1] |
| 141 | function_invoker = ' '.join(function_data[2:-2]) |
| 142 | print('{:>4} {:>10} {:<30} {}'.format(index, |
| 143 | function_percentage, |
| 144 | function_name, |
| 145 | function_invoker)) |
| 146 | |
| 147 | # Remove intermediate files |
| 148 | os.unlink('/tmp/perf.data') |
| 149 | os.unlink('/tmp/perf_report.out') |