On Mon, Dec 18, 2023 at 4:01 PM Nathan Chancellor [off-list ref] wrote:

On Sat, Dec 16, 2023 at 12:53:43PM +0800, kernel test robot wrote:

quoted

Hi Sebastian,

kernel test robot noticed the following build errors:

[auto build test ERROR on net-next/main]

url:    https://github.com/intel-lab-lkp/linux/commits/Sebastian-Andrzej-Siewior/locking-local_lock-Introduce-guard-definition-for-local_lock/20231216-011911
base:   net-next/main
patch link:    https://lore.kernel.org/r/20231215171020.687342-21-bigeasy%40linutronix.de
patch subject: [PATCH net-next 20/24] net: intel: Use nested-BH locking for XDP redirect.
config: arm-defconfig (https://download.01.org/0day-ci/archive/20231216/202312161212.D5tju5i6-lkp@intel.com/config)
compiler: clang version 14.0.6 (https://github.com/llvm/llvm-project.git f28c006a5895fc0e329fe15fead81e37457cb1d1)
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20231216/202312161212.D5tju5i6-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot [off-list ref]
| Closes: https://lore.kernel.org/oe-kbuild-all/202312161212.D5tju5i6-lkp@intel.com/ (local)

All errors (new ones prefixed by >>):

quoted

quoted

drivers/net/ethernet/intel/igb/igb_main.c:8620:3: error: cannot jump from this goto statement to its label

                   goto xdp_out;
                   ^

^ The problematic goto should be replaced with an early return. (and
perhaps a comment that you can't jump over __cleanup variable
initialization).

Otherwise the compiler cannot put the cleanup in the destination basic
block; it would have to split the edges and have all the happy paths
go to a synthesized basic block that runs the cleanup, then jumps to
the original destination.

quoted

   drivers/net/ethernet/intel/igb/igb_main.c:8624:2: note: jump bypasses initialization of variable with __attribute__((cleanup))
           guard(local_lock_nested_bh)(&bpf_run_lock.redirect_lock);
           ^
   include/linux/cleanup.h:142:15: note: expanded from macro 'guard'
           CLASS(_name, __UNIQUE_ID(guard))
                        ^
   include/linux/compiler.h:180:29: note: expanded from macro '__UNIQUE_ID'
   #define __UNIQUE_ID(prefix) __PASTE(__PASTE(__UNIQUE_ID_, prefix), __COUNTER__)
                               ^
   include/linux/compiler_types.h:84:22: note: expanded from macro '__PASTE'
   #define __PASTE(a,b) ___PASTE(a,b)
                        ^
   include/linux/compiler_types.h:83:23: note: expanded from macro '___PASTE'
   #define ___PASTE(a,b) a##b
                         ^
   <scratch space>:52:1: note: expanded from here
   __UNIQUE_ID_guard753
   ^
   1 error generated.

I initially thought that this may have been
https://github.com/ClangBuiltLinux/linux/issues/1886 but asm goto is not
involved here.

This error occurs because jumping over the initialization of a variable
declared with __attribute__((__cleanup__(...))) does not prevent the
clean up function from running as one may expect it to, but could
instead result in the clean up function getting run on uninitialized
memory. A contrived example (see the bottom of the "Output" tabs for the
execution output):

https://godbolt.org/z/9bvGboxvc

While there is a warning from GCC in that example, I don't see one in
the kernel's case. I see there is an open GCC issue around this problem:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91951

While it is possible that there may not actually be a problem with how
the kernel uses __attribute__((__cleanup__(...))) and gotos, I think
clang's behavior is reasonable given the potential footguns that this
construct has.

Cheers,
Nathan

-- 
Thanks,
~Nick Desaulniers