[Linux-kernel-mentees] checkpatch.pl investigation for false-positives

Lukas Bulwahn lukas.bulwahn at gmail.com
Thu Sep 24 17:26:03 UTC 2020

On Wed, 23 Sep 2020, Ujjwal Kumar wrote:

> Hi Lukas,
> I looked at different error messages and came-up with a list of
> possible false positives reports by checkpatch script.
> First, is about BAD_SIGN_OFF warning for Duplicate signature.
> Though it is obvious to have these warnings but I am wondering 
> about situations where there is some other signature between 
> two exactly same signatures.
> For instance,
> commit 3f9c6d38797e9903937b007a341dad0c251765d6
>     ...
>     Signed-off-by: Jonathan Cameron <Jonathan.Cameron at huawei.com>
>     Acked-by: Andrew F. Davis <afd at ti.com>
>     Cc: <Stable at vger.kernel.org>
>     Signed-off-by: Jonathan Cameron <Jonathan.Cameron at huawei.com>
> commit 79ea1e12c0b8540100e89b32afb9f0e6503fad35
>     ...
>     Signed-off-by: Johannes Berg <johannes.berg at intel.com>
>     Link: https://lore.kernel.org/r/20200604120420.b1dc540a7e26.I55
> dcca56bb5bdc5d7ad66a36a0b42afd7034d8be at changeid
>     Signed-off-by: Johannes Berg <johannes.berg at intel.com>
> I'm not familiar how/when these situations can occur. But if these 
> scenarios are commonly acceptable then the script should not report
>  a warning for duplicate signatures.

That is a typical situation when a maintainer sends their own patches to a 
mailing list and then picks them up again. Hmm, I need to think about if 
that should be addressed. You could of course detect the Link: 
with lore.kernel.org and identify that all further sign offs are now part 
of the patch integration. Running checkpatch.pl on git commits is not the
most common thing, so it might just be a minor issue.

Let us keep it in mind for now.

> ---
> Second, is about WHILE_AFTER_BRACE errors.
> Example commit: 6f8a57ccf8511724e6f48d732cb2940889789ab2
> The error message suggests to move the empty while loop with close 
> brace of an if block which is not recommended according to the code 
> style.
> The possible correction would be consider checking the type of 
> block that is ending and report only when the do block is found.

Okay, that is going to be tricky... maybe clang-format and friends is the 
better tool to handle that eventually. I would not pick that to 
investigate further, just because it is quite tricky. checkpatch.pl is 
already just doing some rough estimation and not really parsing the syntax 
> ---
> Third, is about EXECUTE_PERMISSIONS errors where the script reports 
> an error for a python (possible with others scripts as well) script 
> with file mode set as '100755' but file name does not include '.py' 
> extension and is not present under the 'scripts' directory tree.
> So, a python script file named 'binfmt_script' with file mode '100755'
>  is erroneously reported by the script as:
> ERROR:EXECUTE_PERMISSIONS: do not set execute permissions for source files
> #39: FILE: tools/testing/selftests/exec/binfmt_script
> even when it is okay to make the script file executable.
> Example commit b081320f0693cce0394f7c8bad9fba0b25982186
> A possible solution to this would be to use the script's first line 
> for determining whether the file is a executable file or not.

That seems reasonable. Maybe the rule needs an update anyway because 
executable scripts can also be in tools/...

I suggest to have a look at the discussion of this patch where we already 
discussed out the role of execute permissions in the kernel repository:


Mrinal never followed up on the two potential tasks in

Ujjwal, maybe you can look at the build scripts and check if there are any 
places where the build really relies on the file being executable (which 
we currently do guarantee)?

If there are some questions to the task, feel free to ask.


More information about the Linux-kernel-mentees mailing list