smatch devel branch

smatch devel branch

[Dan Carpenter]

I have been re-working Smatch to be useful in an AI world.  I'm making
quite a few changes as quickly as I can.  And quick changes mean is an
element of roughness.

One thing is that I got rid of the check_list.h file.  Now just the
having the check_foo.c file in the smatch/ directory is enough.  No need
to edit anything else.  Hopefully, this sort of thing makes it easier to
write tons and tons of smaller checks.  I really want people to vibe code
checks.  Harshit and I have messed around with that and it's quite
promising.

You can also use AI to filter out false positives and that's quite a nice
combination of automatically writing the check, running it and filtering
out the false positives.

The other thing is that Smatch just has to get better because the
competition is better now.  So for example, I've always thought that
probably parsing the code twice would help.  I held off on doing that
because it makes Smatch use twice as much memory and be twice as slow.
But now I've made that the only option.  I've found other places to
speed up the code and my nightly linux-next builds are completing in
time now.  There are still too many places where Smatch says that
functions are "too hairy".  Before I switched to two pass processing
it was 850 and now it's 3600.  So I need to fix that.

I've got lots and lots of ideas and I don't plan to slow down but
I'm at a place where the code I have so far is almost working and I've
pushed it to the devel branch so people can take a look if they want.

regards,
dan carpenter

Re: smatch devel branch

[Harshit Mogalapalli]

Hi Dan,

Using this thread for few questions.

smatch_kernel_irq_context.c is removed in smatch-devel, but in the past 
we have reports like this: 
https://lore.kernel.org/all/202603252044.Cacz4oRd-lkp@intel.com/

Maybe we should track it ?

Thanks,
Harshit

Re: smatch devel branch

[Harshit Mogalapalli]

Hi Dan,

Regression Report 2 for smatch-devel.

After commit: 76b24bdbac63 ("smatch_data: move data into 
smatch_data/project/ git sub repository") we shouldn't rpbably have 
kernel. prefixes for the files.

but I still see:



../smatch-devel/smatch_data/kernel/kernel.allocation_funcs
../smatch-devel/smatch_data/kernel/kernel.dma_funcs
../smatch-devel/smatch_data/kernel/kernel.frees_argument
../smatch-devel/smatch_data/kernel/kernel.sizeof_param
../smatch-devel/smatch_data/kernel/kernel.unwind_functions


Thanks,
Harshit

Re: smatch devel branch

[Harshit Mogalapalli]

Hi Dan,

Regression report 3:

net/rds/connection.c:547 rds_conn_destroy() warn: mixing irq and irqsave

this is not reproducible  with smatch-devel.

I think some of these are false positives, but anyway wanted to bring 
these to your attention.

Thanks,
Harshit

Re: smatch devel branch

[Dan Carpenter]

On Thu, Jun 11, 2026 at 01:38:38PM +0530, Harshit Mogalapalli wrote:
> Hi Dan,
> 
> Regression report 3:
> 
> net/rds/connection.c:547 rds_conn_destroy() warn: mixing irq and irqsave
> 
> this is not reproducible  with smatch-devel.
> 
> I think some of these are false positives, but anyway wanted to bring these
> to your attention.

Fixed and pushed.  Thanks!

Also that rds_conn_destroy() code is nonsense.  The warning is valid.  :P

I worry there may be other code where I tried to silence the
warning and now nothing is printed...

regards,
dan carpenter