Enhance fastrpc_test for multi-domain and multi-PD execution#406
Conversation
There was a problem hiding this comment.
- changed defaults are not truly backward compatible
- YAML wrapper in the changed hunk still uses the older masked pattern
- the new no-domain skip path exits 1 after writing SKIP
Please address the workflow issues - https://github.com/qualcomm-linux/qcom-linux-testkit/actions/runs/24520486149/job/71764753535?pr=406
smuppand
left a comment
smuppand
left a comment
There was a problem hiding this comment.
Could you please test this in Lava and share the jobs as a reference once the changes are applied?
@bhargav0610 @qcom-anilyada The current run.sh script modifies the default behavior, and fastrpc is designed to change it intentionally. Please take note of this.
anankulk
force-pushed
the
fastrpc-multi-domain-test
branch
3 times, most recently
from
7797d7e to
43a4100
Compare
|
@anankulk Please fix the workflow errors https://github.com/qualcomm-linux/qcom-linux-testkit/actions/runs/24723885995/job/72363028941 |
anankulk
force-pushed
the
fastrpc-multi-domain-test
branch
from
4afc121 to
8a1e3b3
Compare
anankulk
force-pushed
the
fastrpc-multi-domain-test
branch
from
37e4d9f to
2c7b767
Compare
|
@anankulk Fix the workflow issues |
anankulk
force-pushed
the
fastrpc-multi-domain-test
branch
from
2c7b767 to
1a4877f
Compare
|
Acknowledged the comment and updated the patch with the required fix. |
smuppand
left a comment
smuppand
left a comment
There was a problem hiding this comment.
Please do squashing/cleaning the 5 commits
| esac | ||
|
|
||
| pattern="${LOG_ROOT}/${dom_name}_${pd_tag}_iter*.out" | ||
| for iter_file in $pattern; do |
There was a problem hiding this comment.
The final summary still reparses ${LOG_ROOT}/${domain}_${pd}_iter*.out files to compute subtest totals. The earlier request and PR description say results should be tracked during the main run loop instead of reparsing logs. Please either track subtest totals during each iteration and store them in RESULTS_TRACKER, or update the PR description if this reparsing is intentionally retained only for subtest summary.
anankulk
force-pushed
the
fastrpc-multi-domain-test
branch
from
1a4877f to
8b3b28b
Compare
|
Here are the jobs on different targets, please check:
Target | Job | Status -- | -- | -- glymur-crd | https://lava.infra.foundries.io/scheduler/job/224287 | Boot Crash/Timeout iq-8275-evk | https://lava.infra.foundries.io/scheduler/job/224262 | PASS kaanapali-mtp | https://lava.infra.foundries.io/scheduler/job/224266 | FAIL qcs615-adp-air | https://lava.infra.foundries.io/scheduler/job/224285 | PASS qcs6490 | https://lava.infra.foundries.io/scheduler/job/224270 | PASS qcs8300-ride-sx | https://lava.infra.foundries.io/scheduler/job/224276 | FAIL iq-9075-evk | https://lava.infra.foundries.io/scheduler/job/224280 | PASS rb1-core-kit | https://lava.infra.foundries.io/scheduler/job/224281 | SKIP |
anankulk
force-pushed
the
fastrpc-multi-domain-test
branch
from
8b3b28b to
ef36a2d
Compare
|
@anankulk Please fix the workflow issues reported https://github.com/qualcomm-linux/qcom-linux-testkit/actions/runs/26887652006/job/79679040979?pr=406 |
anankulk
force-pushed
the
fastrpc-multi-domain-test
branch
from
af122be to
8c593a0
Compare
|
@anankulk We perform validation on meta-qcom CI first, then proceed with merging to prevent any regressions on other SoCs. |
|
Here are the soak run on this PR on all premerge enabled targets:
|
Thanks @qcom-anilyada for the validation results. @anankulk please check for the qcs9100-ride-sx failure and RB1, Glymur SKIP reasons. |
- Discover DSP domains at runtime via get_remoteproc_by_firmware() with dt_has_remoteproc_fw() as fallback - Support CDSP1, GPDSP0, GPDSP1; canonicalize firmware names - Domain-aware PD execution; rename USER_PD -> UNSIGNED_PD - Per-domain/per-PD tabular summary with aligned subtest counts - SoC blacklist: skip QRB2210/Glymur CRD; filter GPDSP on QCS9075/ QCS8275/QCS8300; skip libhap_example HAP_mem failure on SM8850 - Fix SOC_MACHINE parsing for multi-word SoC names - Fix all exit paths to exit 0 for LAVA .res consumption Signed-off-by: Anand Kulkarni <anankulk@qti.qualcomm.com>
anankulk
force-pushed
the
fastrpc-multi-domain-test
branch
from
8c593a0 to
fb2162f
Compare
|
Thanks for the results @smuppand , @qcom-anilyada. 1)With the new patch, the failing tests on qcs9100-ride-sx (GPDSP) have been temporarily disabled, as fastrpc_tests are currently not supported on GPDSP. We will update the test coverage and re-enable the disabled tests once these issues are resolved. |
|
on the latest commit, here are the soak results, please check:
|
5 participants
Summary
This PR enhances fastrpc_test by expanding test coverage from a single DSP domain to runtime discovery and validation of all supported DSP domains.
The change improves coverage on platforms with multiple DSPs while preserving compatibility with legacy single-domain execution.
Changes
Runtime DSP domain discovery
get_remoteproc_by_firmware()withdt_has_remoteproc_fw()as fallbackcdsp0→cdsp,gdsp0→gpdsp0,gdsp1→gpdsp1)Expanded domain support
Domain-aware PD execution
FASTRPC_UNSIGNED_PDenv var with correct priority:--unsigned-pdCLI >FASTRPC_UNSIGNED_PDenv >--pd-modeResult tracking and reporting
RESULTS_TRACKERduring run loopSoC-specific handling
QRB2210andGlymur CRDentirely — FastRPC not supportedGPDSP0/GPDSP1(domains 5/6) forQCS9075,QCS8275andQCS8300libhap_exampleHAP_mem DMA failure as known skip onSM8850— remaining subtests passSOC_MACHINEextraction to preserve full string for reliable multi-word SoC name matchingBehavior
Default behavior
Legacy behavior
--domain-mode single --domain <N>Maintenance
pick_default_domain()helperUSER_PD→UNSIGNED_PDacross YAML params, CLI flags and env varsrun.shusage documentationExpected Sample Output: (qcs9100-ride-sx)
