perf tests: top/kvm don't depend on coreutils timeout
Ian Rogers
reviewerPATCH 1/1 perf tests: top/kvm don't depend on coreutils timeout
If the coreutils timeout command is missing on a minimal test system then the kvm and top tests can fail. Provide a fallback bash implementation for these systems.
--- tools/perf/tests/shell/kvm.sh | 6 ++++- tools/perf/tests/shell/lib/timeout.sh | 36 +++++++++++++++++++++++++++ tools/perf/tests/shell/top.sh | 7 +++++- 3 files changed, 47 insertions(+), 2 deletions(-) create mode 100644 tools/perf/tests/shell/lib/timeout.sh
tools/perf/tests/shell/kvm.sh +5 -1
@@ -4,6 +4,9 @@ set -e +# shellcheck source=lib/timeout.sh +. "$(dirname $0)"/lib/timeout.sh + err=0 perfdata=$(mktemp /tmp/__perf_kvm_test.perf.data.XXXXX) qemu_pid_file=$(mktemp /tmp/__perf_kvm_test.qemu.pid.XXXXX)
@@ -102,7 +105,8 @@ test_kvm_stat_live() { # Run perf kvm live for 5 seconds, monitoring that PID # Use sleep to keep stdin open but silent, preventing EOF loop or interactive spam - if ! sleep 10 | timeout 5s perf kvm stat live -p "${qemu_pid}" > "${log_file}" 2>&1; then + if ! sleep 10 | bash_timeout 5s perf kvm stat live -p "${qemu_pid}" > "${log_file}" 2>&1 + then retval=$? if [ $retval -ne 124 ] && [ $retval -ne 0 ]; then echo "perf kvm stat live [Failed: perf kvm stat live failed to start or run (ret=$retval)]"
tools/perf/tests/shell/lib/timeout.sh new file +36 -0
@@ -0,0 +1,36 @@ +#!/bin/bash +# SPDX-License-Identifier: GPL-2.0 + +# A version of timeout in pure bash that doesn't require the timeout command. +bash_timeout() { + # Check if the native 'timeout' command is available + if command -v timeout >/dev/null 2>&1; then + timeout "$@" + return $? + fi + + local timeout_secs="$1" + shift + local command=("$@") + + # Execute the target command in the background + "${command[@]}" & + local cmd_pid=$! + + # Start a watcher process in the background + ( + sleep "$timeout_secs" + kill -TERM "$cmd_pid" 2>/dev/null || true + ) & + local watcher_pid=$! + + # Wait for the main command to complete (or be killed) + wait "$cmd_pid" 2>/dev/null + local exit_code=$? + + # Clean up: Kill the watcher process if the main command finished early + kill "$watcher_pid" 2>/dev/null || true + + # Return the exit status of the target command + return $exit_code +}
tools/perf/tests/shell/top.sh +6 -1
@@ -4,6 +4,9 @@ set -e +# shellcheck source=lib/timeout.sh +. "$(dirname $0)"/lib/timeout.sh + err=0 log_file=$(mktemp /tmp/perf.top.log.XXXXX)
@@ -35,7 +38,9 @@ test_basic_perf_top() { # Use -d 1 to avoid flooding output # Use -e cpu-clock to ensure we get samples # Use sleep to keep stdin open but silent, preventing EOF loop or interactive spam - if ! sleep 10 | timeout 5s perf top --stdio -d 1 -e cpu-clock -p $PID > "${log_file}" 2>&1; then + if ! sleep 10 | \ + bash_timeout 5s perf top --stdio -d 1 -e cpu-clock -p $PID > "${log_file}" 2>&1 + then retval=$? if [ $retval -ne 124 ] && [ $retval -ne 0 ]; then echo "Basic perf top test [Failed: perf top failed to start or run (ret=$retval)]"
Signature · 1 line
2.54.0.rc0.605.g598a273b03-goog
[email protected]
AI reviewSashiko AI review found 4 potential issue(s): - [High] Orphaned watcher process can send SIGTERM to unrelated system processes (PID recycling). - [Medium] Exit code mismatch in timeout emulation…
Sashiko AI review found 4 potential issue(s):
- [High] Orphaned watcher process can send SIGTERM to unrelated system processes (PID recycling).
- [Medium] Exit code mismatch in timeout emulation falsely fails tests.
- [Medium] Implicit standard input redirection breaks the protective pipeline.
- [Medium] Non-POSIX sleep argument causes immediate timeout on strict minimal systems. --
commit 17f50e3b3f4410cb65917a2951f5c28c36d9bf25
Author: Ian Rogers <[email protected]>
perf tests: top/kvm don't depend on coreutils timeout
This commit adds a bash-based fallback for the coreutils timeout command to prevent failures on minimal test systems. It then updates the kvm and top test scripts to use this new bash_timeout function.
Quoted 4 lines
tools/perf/tests/shell/kvm.sh +0
-0
[ ... ]
tools/perf/tests/shell/kvm.sh test_kvm_stat_live() { · line 108
Quoted 6 lines
tools/perf/tests/shell/kvm.sh +1
-1@@ -102,7 +105,8 @@ test_kvm_stat_live() { # Run perf kvm live for 5 seconds, monitoring that PID # Use sleep to keep stdin open but silent, preventing EOF loop or interactive spam - if ! sleep 10 | timeout 5s perf kvm stat live -p "${qemu_pid}" > "${log_file}" 2>&1; then + if ! sleep 10 | bash_timeout 5s perf kvm stat live -p "${qemu_pid}" > "${log_file}" 2>&1
Does POSIX sleep support the "s" suffix passed as "5s" here?
On minimal systems lacking coreutils, sleep "5s" might fail instantly, causing the watcher script to kill the target command in 0 seconds.
tools/perf/tests/shell/kvm.sh test_kvm_stat_live() { · line 111
Quoted 3 lines
tools/perf/tests/shell/kvm.sh +1
-0+ then retval=$? if [ $retval -ne 124 ] && [ $retval -ne 0 ]; then
When bash_timeout kills the process via SIGTERM, it returns 143 (128 + 15), not the GNU timeout exit code of 124.
Since this code checks for 124, will an expected timeout incorrectly register as a test failure?
Quoted 5 lines
tools/perf/tests/shell/lib/timeout.sh new file +0
-0
[ ... ]
Quoted 1 line
+bash_timeout() {
[ ... ]
tools/perf/tests/shell/lib/timeout.sh line 17
Quoted 6 lines
tools/perf/tests/shell/lib/timeout.sh new file +6
-0+ local timeout_secs="$1" + shift + local command=("$@") + + # Execute the target command in the background + "${command[@]}" &
Does placing the target command in the background automatically redirect its standard input from /dev/null?
This appears to sever the connection to the "sleep 10 |" pipe used in the calling scripts, causing perf to read an immediate EOF and triggering the loop and interactive spam the pipe was meant to prevent.
tools/perf/tests/shell/lib/timeout.sh line 24
Quoted 7 lines
tools/perf/tests/shell/lib/timeout.sh new file +7
-0+ local cmd_pid=$! + + # Start a watcher process in the background + ( + sleep "$timeout_secs" + kill -TERM "$cmd_pid" 2>/dev/null || true + ) &
If the test script exits early (for example, due to set -e or an interrupt) before killing the watcher process, does the watcher survive as an orphaned background task?
Since perf tests often run as root, this orphaned watcher might wake up later and send SIGTERM to a recycled PID, potentially terminating unrelated system services.
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1