drivers/infiniband/hw/qib/qib_sysfs.c:413:1: error: static_assert expression is not an integral constant expression

[nathanchance]

After https://git.kernel.org/linus/84f969e1c48ed3825986e91a0786e363d57f69d1, which is now in mainline:

$ make -skj"$(nproc)" LLVM=1 allmodconfig drivers/infiniband/hw/qib/qib_sysfs.o
drivers/infiniband/hw/qib/qib_sysfs.c:413:1: error: static_assert expression is not an integral constant expression
QIB_DIAGC_ATTR(rc_resends);
^~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/infiniband/hw/qib/qib_sysfs.c:406:16: note: expanded from macro 'QIB_DIAGC_ATTR'
        static_assert(&((struct qib_ibport *)0)->rvp.n_##N != (u64 *)NULL);    \
        ~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
./include/linux/build_bug.h:77:50: note: expanded from macro 'static_assert'
#define static_assert(expr, ...) __static_assert(expr, ##__VA_ARGS__, #expr)
                                 ~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~
./include/linux/build_bug.h:78:56: note: expanded from macro '__static_assert'
#define __static_assert(expr, msg, ...) _Static_assert(expr, msg)
                                                       ^~~~
drivers/infiniband/hw/qib/qib_sysfs.c:413:1: note: cannot access field of null pointer
drivers/infiniband/hw/qib/qib_sysfs.c:406:43: note: expanded from macro 'QIB_DIAGC_ATTR'
        static_assert(&((struct qib_ibport *)0)->rvp.n_##N != (u64 *)NULL);    \
                                                 ^
...

GCC back to at least 7.5.0 has no issues with this construct.

[nathanchance]

I am guessing the note: cannot access field of null pointer is the crux of the matter.

[nickdesaulniers]

Should this be changed to BUILD_BUG_ON instead of using static_assert?

[nathanchance]

I'll test it to see if it fixes it.

[nathanchance]

Ah, BUILD_BUG_ON won't work as these structures are not declared within a function but globally within the file so the do { ... } while (0) from __compiletime_assert does not work.

[nickdesaulniers]

I don't understand the check; they're testing the offset of a member isn't 0? Sounds like they should be using offset of, not pointer comparisons.

[nathanchance]

They do use offsetof. Are they making sure the member of the structure is the size of a u64 *?

[nickdesaulniers]

Upstream thread: https://lore.kernel.org/llvm/20210913165317.GX2505917@nvidia.com/.

[dileks]

I can confirm this issue with Linux v5.15-rc1+ and LLVM v13.0.0-rc3.
Happily, it is the only error breaking my 1st build.

[nathanchance]

We have a solution pending: https://lore.kernel.org/r/0-v1-1b789bd4dbd4+14b16-clang-fix_jgg@nvidia.com/

[dileks]

Thanks.
That is what I applied as a diff and was able to continue building.

$ cat /proc/version 
Linux version 5.15.0-rc1-1-amd64-clang13-lto (sedat.dilek@gmail.com@iniza) (dileks clang version 13.0.0 (https://github.com/llvm/llvm-project.git 1f27fe6128769f00197925c3b8f6abb9d0e5cd2e), LLD 13.0.0 (https://github.com/llvm/llvm-project.git 1f27fe6128769f00197925c3b8f6abb9d0e5cd2e)) #1~bookworm+dileks1 SMP 2021-09-15

[dileks]

Patch is in rdma Git tree.

[1] https://git.kernel.org/rdma/rdma/c/1b789bd4dbd48a92f5427d9c37a72a8f6ca17754

Update 2021-09-23: Commit-ID changed in rdma.git#for-rc.

[1] https://git.kernel.org/rdma/rdma/c/3110b942d36b961858664486d72f815d78c956c3

[dileks]

Now in Linus Git.

[1] https://git.kernel.org/linus/3110b942d36b961858664486d72f815d78c956c3