Klever

including 3.14

Incorrect. In >3.12 macro DECLARE_COMPLETION_ONSTACK() is processed as INIT_COMPLETION(), whereas for more earlier kernels those macros are different.

For kernels ~2.6.33 fixed in branch fix_148_2.6.33.

I didn't noticed any changes in corresponding API. What differs and matters is a configuration. Before your fix there is just a model that works when CONFIG_LOCKDEP is true (which is usually the case for us). But when it is false, there isn't a model.

So first macro DECLARE_COMPLETION_ONSTACK should be modeled as you did. Second there should be a test case with an appropriate configuration that detects this issue.

Vitaly Mordan wrote:

Incorrect. In >3.12 macro DECLARE_COMPLETION_ONSTACK() is processed as INIT_COMPLETION()

My mistake, it is not connected to kernel version, but to its configuration.
With allmodconfig this issue cannot be revealed even in 2.6.33.

Confirmed, that in branch fix_148_2.6.33 all related false positive have gone.

There isn't still an appropriate test case that can reveal this issue. So I am going to add it myself.

There should be new job with another configuration for that test case.

Vitaly Mordan wrote:

There should be new job with another configuration for that test case.

Obviously, but you didn't add a corresponding sub-job.

Adding 2 sub-jobs for one module/rule leads to #7917. This should be fixed.