performance: add LAVA-ready YAMLs and env/CLI alignment for Boot_Systemd_Validate + KPI_Loop #237
Conversation
smuppand
force-pushed
the
systemd-boot
branch
from
12c4529 to
99624c7
Compare
smuppand
force-pushed
the
systemd-boot
branch
2 times, most recently
from
a396de2 to
d26c52c
Compare
| # Put all CPUs into performance governor, saving previous governor for restore. | ||
| # Uses SAVED_GOV_FILE (auto set if not provided). | ||
| set_performance_governor() { | ||
| SAVED_GOV_FILE="${SAVED_GOV_FILE:-/tmp/perf_saved_governors.$$}" |
There was a problem hiding this comment.
What if its already set to performance? should we check that first and return if its already set?
There was a problem hiding this comment.
What if its already set to performance? should we check that first and return if its already set?
QLI uses the default arm64 debug defconfig, which does not set the performance governor by default.
|
|
||
| # Put all CPUs into performance governor, saving previous governor for restore. | ||
| # Uses SAVED_GOV_FILE (auto set if not provided). | ||
| set_performance_governor() { |
There was a problem hiding this comment.
Can this fnction be made generic to userspace/shedutil aswell?
There was a problem hiding this comment.
Can this fnction be made generic to userspace/shedutil aswell?
Wht do you mean generic? It is already generic, and run.sh can import these functions by calling lib_performance.sh.
There was a problem hiding this comment.
i meant to have a function which will set to performance/userspace or shedutil based on the argument instead of having for performance. Let me know your thought?
There was a problem hiding this comment.
i meant to have a function which will set to performance/userspace or shedutil based on the argument instead of having for performance. Let me know your thought?
We can handle this in a separate PR. You can create a GitHub issue ticket for this requirement.
Runner/utils/lib_performance.sh
Outdated
| if [ "$disable_getty" = "1" ]; then | ||
| systemctl disable serial-getty@ttyS0.service >/dev/null 2>&1 || true | ||
| systemctl stop serial-getty@ttyS0.service >/dev/null 2>&1 || true | ||
| if command -v log_info >/dev/null 2>&1; then |
There was a problem hiding this comment.
can this be removed to optimize script as log_warn/log_info will always be there and if not execution wont reach till here
Runner/utils/lib_performance.sh
Outdated
| timeout="$1" | ||
|
|
||
| if ! command -v systemctl >/dev/null 2>&1; then | ||
| if command -v log_warn >/dev/null 2>&1; then |
There was a problem hiding this comment.
can this be removed to optimize script as log_warn/log_info will always be there and if not execution wont reach till here
Runner/utils/lib_performance.sh
Outdated
| ts=$(nowstamp) | ||
| echo "$ts,$bt,$PERF_KPI_ITERATIONS_HINT,$PERF_KPI_CLOCKSOURCE,$PERF_KPI_UEFI_TIME_SEC,$PERF_KPI_FIRMWARE_SEC,$PERF_KPI_BOOTLOADER_SEC,$PERF_KPI_KERNEL_SEC,$PERF_KPI_USERSPACE_SEC,$PERF_KPI_USERSPACE_EFFECTIVE_SEC,$PERF_KPI_BOOT_TOTAL_SEC,$PERF_KPI_BOOT_TOTAL_EFFECTIVE_SEC" >>"$csv" 2>/dev/null || true | ||
|
|
||
| if command -v log_info >/dev/null 2>&1; then |
There was a problem hiding this comment.
can this be removed to optimize script as log_warn/log_info will always be there and if not execution wont reach till here
Runner/utils/lib_performance.sh
Outdated
| perf_kpi_request_reboot() { | ||
| msg=$1 | ||
|
|
||
| if command -v log_info >/dev/null 2>&1; then |
There was a problem hiding this comment.
can this be removed to optimize script as log_warn/log_info will always be there and if not execution wont reach till here
|
|
||
| # If we are still alive after a short delay, try once more. | ||
| sleep 5 | ||
| if command -v systemctl >/dev/null 2>&1; then |
There was a problem hiding this comment.
Should we try to run in loop with timeout/iteration instead?
There was a problem hiding this comment.
Should we try to run in loop with timeout/iteration instead?
This function is not completely optimized since full optimization isn't necessary for Boot_Systemd_Validate.
Runner/utils/lib_performance.sh
Outdated
| fi | ||
|
|
||
| # Boot-id changed → reboot successful, just log it and clear pending flag in state. | ||
| if command -v log_info >/dev/null 2>&1; then |
There was a problem hiding this comment.
can this be removed to optimize script as log_warn/log_info will always be there and if not execution wont reach till here
| log_warn "Unknown option: $1" | ||
| usage >&2 | ||
| echo "$TESTNAME FAIL" >"$RES_FILE" | ||
| exit 1 |
There was a problem hiding this comment.
Should we use exit 0 only?
| mkdir -p "$OUT_DIR" || { | ||
| log_error "Cannot create $OUT_DIR" | ||
| echo "$TESTNAME FAIL" >"$RES_FILE" | ||
| exit 1 |
|
|
||
| # --- ensure CPU governors restored on exit --- | ||
| cleanup() { | ||
| restore_governor |
There was a problem hiding this comment.
Check if this is reallyrequired as we are not setting the performance governer above
|
|
||
| BOOT_NOT_FINISHED=0 | ||
|
|
||
| if command -v systemd-analyze >/dev/null 2>&1; then |
There was a problem hiding this comment.
this is already checked in check_dependencies and test will be skipped if not present. So this check can be removed
There was a problem hiding this comment.
There’s no harm in checking since it won’t be redundant.
| fi | ||
|
|
||
| # ---------- Failed units + journal ---------- | ||
| if command -v systemctl >/dev/null 2>&1; then |
There was a problem hiding this comment.
i see there is a lot of systemctl dependency in the script. why not check in check_dependency and skip the test if not available?
There was a problem hiding this comment.
There’s no harm in checking since it won’t be redundant.
| # ---------- required units gating ---------- | ||
| suite_rc=0 | ||
| if [ -n "$REQ_UNITS_FILE" ]; then | ||
| if command -v systemctl >/dev/null 2>&1; then |
There was a problem hiding this comment.
Move to check_dependencies and remove here
Expose wait_analyze_ready() to poll systemd-analyze time safely. Record systemctl list-jobs whenever boot is not yet finished. Allow configurable timeout and poll interval via environment variables. Prepare shared helpers for reuse across performance KPI test suites. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>
- Replace the ad-hoc systemd-analyze polling loop with the shared helper. - Capture list-jobs when boot is unfinished and log this explicitly. - Honor extended boot times on slow platforms before parsing KPIs. - Preserve existing outputs while avoiding misleading 'boot not finished' results. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>
smuppand
force-pushed
the
systemd-boot
branch
from
d26c52c to
f4fdfb0
Compare
Switch auto-reboot orchestration to a oneshot systemd service + timer. Let the loop manage iterations/state while keeping the boot path clean. Ensure the KPI service exits quickly so FinishTimestampMonotonic is not blocked. Document usage while preserving behavior for manual single-shot runs. Signed-off-by: Srikanth Muppandam <smuppand@qti.qualcomm.com>
smuppand
force-pushed
the
systemd-boot
branch
from
f4fdfb0 to
4046117
Compare
2 participants
This PR updates the Boot Systemd KPI tests to be fully LAVA-friendly by delivering complete, configurable run.sh + YAML test definitions, and aligning CLI options ↔ LAVA variables/env without changing core KPI logic.
Included tests