Update domain tests for RFC-20

[Indanz]

Test with seL4/seL4#1511.

[lsf37]

The tests looks good to me, any idea why we are getting the PC99_debug_DOM_clang_32 simulation failure?

The end of the log is this:

  	<testcase classname="sel4test" name="DOMAINS0001">
  
  Running test DOMAINS0001 (Change domain successfully())
  
  +++ python3 ../projects/seL4_libs/libsel4test/tools/extract_results.py -q results.xml parsed_results.xml
  Failed to strip leading and trailing garbage

which could be a timeout, although I would have hoped that I'd put a timeout message in the script if it stopped because of one.

The same test does not seem to time out for other platforms and it should be platform independent. Maybe the default 1GHz assumption is not right for qemu?

[Indanz]

which could be a timeout, although I would have hoped that I'd put a timeout message in the script if it stopped because of one.

No, the old tests don't test anything. You're supposed to enable debug prints and check that the result is what you'd expect.

I'm busy improving them and turning them into proper tests.

That said, detecting hangs would complicate individual tests which currently are very simple. (In this case, just two seL4_DomainSet_Set calls.)

Best way to handle timeouts is to have generic test timeout handling in the upper test layer. But considering there isn't even a generic fault handler and how wanting the test infrastructure is, I'm not keen on wasting too much time on this (I'd rather spend my time on rewriting the whole thing personally, to reduce over all complexity). Better to reduce the timeouts in CI for now.

The same test does not seem to time out for other platforms and it should be platform independent. Maybe the default 1GHz assumption is not right for qemu?

That test doesn't use those. It was caused by the ksDomScheduleIdx = domScheduleLength - 1; instead of -2.

[lsf37]

(Still squashing and style update to do, but I'm happy with the content)

[Indanz]

Maybe trivial, but I think you meant to say CMake style instead of camke style. Too late to change though. :-)

[lsf37]

Indeed I did. Sorry about that.

[kent-mcleod]

This looks weird? Are both of these lines still needed?

[Indanz]

Not sure, can't remember whether it was just undoing any changes the test does, or for a specific reason.

I also don't know whether sel4test creates a new task for each tests or not. If it does, then it's redundant for sure.

[kent-mcleod]

I mean that the added seL4_DomainSet_Set() function call comes directly after another seL4_DomainSet_Set() function call to the same tcb. So the previous call looks like it should be removed?

I think this line should either be removed, or a comment added explaining why the the cleanup needs to set the domain first to CONFIG_NUM_DOMAINS - 1 and then again to 0 for the same thread.

[lsf37]

Is this not an actual test? It is setting the domain of the current thread and then resetting it. The reset might not be needed if the thread is suspended anyway afterwards, but it does look cleaner to me to leave it in the domain it started in.

[Indanz]

Yes, the first line is the actual test, hence we store the result in error.
The second line is the cleanup.

[kent-mcleod]

ohhh. The way that GitHub ux was presenting the change it looked like this was a test cleanup function...