I have begun to run Smatch on the U-Boot source code. This is showing up
a number of different classes of issue.
1) Inconsistent indentation.
While I would like to fix all of these reports, they are not actually
causing any bugs. So I am going to assume that in the interest of
keeping churn in the source down to a minimum that these can be ignored.
2) Warnings about large usage of the stack.
I am going to ignore these, at least for the moment. It would be good
practice not to do this though.
3) Uninitialised variables.
These generally result from possible but unlikely code paths. These
should be checked though.
4) Possible buffer overflows.
Most of these are probably not possible in practice but Smatch has not
been able to absolutely confirm that. Some however are going to be real
bugs. These need to be checked.
5) Ignoring unreachable code.
Mostly this is deliberate (if unnecessary). Sometimes it is the result
of config choices. Also used as a way to disable some broken test code.
6) Variable dereferenced before check.
There are instances where a variable is dereferenced early in the
function and then later there is a null check for that variable.
7) Always true or always false tests.
eg checking if an unsigned variable is negative.
8) Probable incorrect tests.
eg use of && instead of &, etc
9) Other things.
I am going to follow this email with a sample email that can be produced
for one report that I believe shows an actual bug.
I do not wish to swamp the mailing list but I do also believe that we
should be aware of these issues and absolutely must resolve them where
needed.
The code coverage from a run of Smatch depends on the platform chosen as
the target of the build and I am never going to run all of them. But
maybe it will be possible to run at least one build in CI with Smatch
enabled to catch newly introduced issues, at least in those areas of
code that are used in that build. It would be good for all developers to
run Smatch on their platform builds to broaden the code coverage as far
as possible.
Andrew
