Klever

For kernels ~2.6.33 fixed in branch fix_153_2.6.33.

I still don't observe a test case so the issue wasn't fixed.

Tests are executed in 3.14, this problem can not be found there. Test will be pointless.

Tests can be executed on any version of the Linux kernel but I expect you will add a test case for Linux 3.14 where this issue does exist. You should inspect a list of potential false alarms here (the concrete example).

The model implemented is incorrect, since usually DEFINE_IDR is used in the global scope, while you modeled it as a function call. The test case is incorrect because of you mixed declarations (DEFINE_IDR) and expressions. Most likely you didn't even try your model on real Linux kernel modules with false alarms.

The real problem is that klever does not support static initialization by such macros, as old LDV Tools did:

[% FOREACH sign = static_inits; PROCESS make_human_sign %]

One more degradation of klever.

I changed rule linux:idr by allowing multiple inits.
Now there is another false positive (module drivers/scsi/sg.ko), it looks like because of EMG.

I don't see any changes in this branch in the repository. Also I hope that you fix and try the developed test case.

508d6280

Vitaly Mordan wrote:

508d6280

Likely you pushed your changes into an incorrect repository.

Commited again.
Rule linux:idr and its tests are adjusted.

Of course, you made it incorrect.
If you initialize each idr with

ldv_idr_sg_index_idr = LDV_INITIALIZED

Evgeny Novikov wrote:

I made everything required in branch fix-linux-lib-idr. Correct tests are passed. The false alarm for Linux 2.6.33 has gone. But instead 4 new interesting unsafes were reported. Ilja Zakharov will review and print statistics here. After that we will decide what to do further.

All those unsafes are false alarms.