Re: sparse: Lots of (erroneous?) "different address space" errrors in new version

[Dan Carpenter <dan.carpenter@linaro.org>]

On Mon, Nov 24, 2025 at 09:55:18AM +0100, Ricardo Ribalda wrote:
> Hi Dan
> 
> Every week we bump the smatch/sparse version used by Linux-media's CI.
> 
> We were almost error free, but the last version has introduced a huge
> amount of errors:
> https://linux-media.pages.freedesktop.org/-/users/ci/-/jobs/88441453/artifacts/junit/test-sparse.log.txt
> 
> (It is also reproducible with linus/master)
> 
> 
> I have bisected your tool and it seems like they were introduced by:
> commit e53027a4e816a772403baafa83c09e4a94c1cb8f (HEAD)
> Author: Dan Carpenter <dan.carpenter@linaro.org>
> Date:   Sat Oct 18 12:31:01 2025 +0300
> 
>     symbol: preserver address space qualifiers with typeof()
> 
>     When we're parsing typeof(var) we should preserve the address space
>     qualifiers as well.
> 
>     Reported-by: Uros Bizjak <ubizjak@gmail.com>
>     Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
> 
> 
> One example of the new error found is:
> 
> make -i W=1 C=1 CHECK="sparse" CF="-D__CHECK_ENDIAN__
> -fmemcpy-max-count=11000000" KCFLAGS="-Wmaybe-uninitialized"
> drivers/media/v4l2-core/v4l2-compat-ioctl32.o
> drivers/media/v4l2-core/v4l2-compat-ioctl32.c:763:13: warning:
> incorrect type in argument 2 (different address spaces)
> 09:08 drivers/media/v4l2-core/v4l2-compat-ioctl32.c:763:13: expected
> void const *from
> 09:08 drivers/media/v4l2-core/v4l2-compat-ioctl32.c:763:13: got
> unsigned int __user *
> 
> 
> static int put_v4l2_event32_time32(struct v4l2_event *p64,
>                                    struct v4l2_event32_time32 __user *p32)
> {
>         if (put_user(p64->type, &p32->type)
> 
> 
> Seems like if we would like to fix the error, put_user should be modified....
> 
> Maybe it would be better to revert `symbol: preserver address space
> qualifiers with typeof()` until the codebase is ready for it?

Huh.  Sorry, I don't know how I missed your email last week.

The do_put_user_call() is doing this:

	__typeof__(*(ptr)) __x = (x); /* eval x once */

And ptr is the __user pointer so Sparse is doing the correct thing.
We could change the __typeof__ to __typeof_unqual__.  It silences the
warning.  I'll rebuild the entire kernel to see if it causes any other
issues...

regards,
dan carpenter