rfmvh / perftool-testsuite Goto Github PK

We already have some coverage for:
event_analyzing_sample, syscall-counts and wakeup-latency. Please. cover the remaining scripts.

For more information, see upstream linux commit 6ef81c55a2b6584cb642917f5fdf3632ef44b670. In short, add a testcase that verifies that perf c2c report -input returns non-zero exit code. This could be probably somehow added into base_c2c/test_basic.sh.

The --switch-output command line argument for perf record allows some external control over the perf recording session. Such a test might contain a crafted tracee, that would signal perf with SIGUSR and check the perf.data* results after, i.e.

signal_perf();
do_some_load();
signal_perf();
do_some_other_load();
signal_perf();

If do_some_load() would be e.g. one third of do_some_other_load(), we might compare the records and verify that the second one brings more samples, etc. Please add this as a separate test script under base_record. TIA.

Currently the perftools testsuite depends on PATH env to find perf command.
However, most possible use case of tests is developing phase, like adding new feature or fixing some bugs. In that case, we usually don't install until it can pass the tests.
Thus, the testsuites must run against pre-installed binaries.

So, please add an option to pass PERF_PATH or something like that to make file or run_all command so that user can give the development kernel tree, e.g.

# make PERF_PATH=$HOME/linux/tools/perf/ check

For more information, please refer to upstream linux commit 8af19d66b956401bab1ef24049eec9421be93862. In short, perf report --header takes information about the date/time of the measurement/sampling from the perf.data file timestamp. Before the patch, it used last-modification-time, so any filesystem operation with the perf.data file clobbered the correct timestamp. Now we use creation-time, so activities as tar/untar should not affect it. Add a testcase to check that the timestamp is still the same even after tar/untar-ing the perf.data file.

Create some tests for perf buildid-cache command and its functionality.

How to run individual set of tests with out using run_all.
My main intention is to run a particular set of tests,
Example:
./run base_mem

I don't see any option with make either.

Thanks in advance.