Age | Commit message (Collapse) | Author |
|
It's better to do any performance measurement in post processing. This
keeps the test scripts simpler and splits the tasks clearly.
While at it, remove the unused cyclictest format parsing from
parse_rt_tests_results since it is not used anymore.
Signed-off-by: Daniel Wagner <wagi@suse.de>
|
|
Start a background process to create load on the system while running
the test.
Signed-off-by: Daniel Wagner <wagi@monom.org>
|
|
cyclictest output is meant for interactive terminals. On dumb
terminals, it creates lots of noise resulting in massive logs.
Use quiet output (-q) to silence and just print a summary at the end
of the test.
Signed-off-by: Kevin Hilman <khilman@baylibre.com>
|
|
Signed-off-by: Milosz Wasilewski <milosz.wasilewski@linaro.org>
|
|
Update all rt-tests with the current upstream v1.4 release.
The x86_64 and armeabi architecture binaries are build using a Debian
stretch root filesystem.
arm64 binaries were build using a Debian buster root filesystem.
The command to build the binaries is: "CLFAGS=-static make"
Signed-off-by: Daniel Wagner <wagi@monom.org>
|
|
The nanosleep option has been removed from cyclictest in commit
9d2c9edd8694 ("cyclictest: make clock_nanosleep as default and add
option for POSIX timers.").
Signed-off-by: Daniel Wagner <wagi@monom.org>
|
|
Add duration command line to cyclictest. rt-tests v1.4 added to many
tools the option '-D'/'--duration'.
This makes the math how long the program is suppose to run
max with --loops and --interval far simpler.
Remove the loops argument in favor of duration.
While at it also add the missing affinity option in the usage output.
Signed-off-by: Daniel Wagner <wagi@monom.org>
|
|
The test should fail if the max value is larger then the expected
MAX_LATENCY has been observed.
MAX_LATENCY value is board and kernel specific, which means the
job needs to set it accordingly.
Unfortunatly, we need to add a slightly more complex parser for
finding the numbers and add the pass/fail flag.
Signed-off-by: Daniel Wagner <wagi@monom.org>
|
|
Don't run cyclictest at priority 99 because the highest priority is
for the high priority kernel tasks. The highest user land priority
is 98.
Furthermore, we should make sure the threads are not migrated (set the
affinity). Also all memory should preallocated (mlockall) before
starting the measurement.
Additional the recommended why to sleep is to use clock_nanosleep
which is not default.
And finally also set the interval to 1000us which most -rt maintainers
are using. This allows to compare the result easier.
Signed-off-by: Daniel Wagner <wagi@monom.org>
|
|
In my case, my system has already get cyclictest installed, so just use
it.
Signed-off-by: Lei Yang <Lei.Yang@windriver.com>
Signed-off-by: Chase Qi <chase.qi@linaro.org>
|
|
Other tests from rt-tests already have this tag. With this patch, we
will be able to filter them easily.
Change-Id: Ic1e62ffe4680a09da73f5e62d136e82a79b5d37e
Signed-off-by: Chase Qi <chase.qi@linaro.org>
|
|
Change-Id: I10181a7d1ebc53491085d80d67f0cd8f9e43071d
Signed-off-by: Daniel Díaz <daniel.diaz@linaro.org>
|
|
Change-Id: I2a81ff35db40935fe57d25bc35685cf6b54c1fe6
Signed-off-by: Chase Qi <chase.qi@linaro.org>
|