linux-kernel - Re: [PATCH] treewide: Remove unnamed static initializations to 0

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 10 Sep 2021 19:08:28 -0700
From:   Kees Cook <keescook@...omium.org>
To:     Alexei Starovoitov <alexei.starovoitov@...il.com>
Cc:     Linus Torvalds <torvalds@...ux-foundation.org>,
        linux-kernel@...r.kernel.org, bpf@...r.kernel.org,
        daniel@...earbox.net, andrii@...nel.org
Subject: Re: [PATCH] treewide: Remove unnamed static initializations to 0

On Fri, Sep 10, 2021 at 04:23:03PM -0700, Alexei Starovoitov wrote:
> On Fri, Sep 10, 2021 at 03:52:07PM -0700, Kees Cook wrote:
> > GCC 4.9 does not like having struct assignments to 0 when members may be
> > compound types. For example, there are 186 instances of these kinds of
> > errors:
> > 
> > drivers/virtio/virtio_vdpa.c:146:9: error: missing braces around initializer [-Werror=missing-braces ]
> > drivers/cxl/core/regs.c:40:17: error: missing braces around initializer [-Werror=missing-braces]
> > 
> > Since "= { 0 }" and "= { }" have the same meaning ("incomplete
> > initializer") they will both initialize the given variable to zero
> > (modulo padding games).
> > 
> > After this change, I can almost build the "allmodconfig" target with
> > GCC 4.9 again.
> > 
> > Signed-off-by: Kees Cook <keescook@...omium.org>
> 
> ...
> 
> >  .../selftests/bpf/prog_tests/perf_branches.c  |   4 +-
> >  .../selftests/bpf/prog_tests/sk_lookup.c      |  12 +-
> >  .../selftests/bpf/prog_tests/sockmap_ktls.c   |   2 +-
> >  .../selftests/bpf/prog_tests/sockmap_listen.c |   4 +-
> >  .../selftests/bpf/progs/test_sk_assign.c      |   6 +-
> >  .../selftests/bpf/progs/test_xdp_vlan.c       |   8 +-
> 
> Those have nothing to do with GCC. They are compiled with clang with -target bpf.
> Did you check that bpf selftests still pass?
> We've had issues with older clang generating different code with zero and non-zero
> assignments and libbpf was confused.
> It should all work now, but please run the tests.

Sure! I think selftests/bpf/config is missing:

CONFIG_DEBUG_INFO=y
CONFIG_DEBUG_INFO_BTF=y

I can't get much further, though:

$ make -C tools/testing/selftests gen_tar TARGETS="bpf" FORMAT=.xz
make: Entering directory '/srv/code/tools/testing/selftests'
make --no-builtin-rules ARCH=x86 -C ../../.. headers_install
make[1]: Entering directory '/srv/code'
  INSTALL ./usr/include
make[1]: Leaving directory '/srv/code'
make: *** [Makefile:162: all] Error 1
make: Leaving directory '/srv/code/tools/testing/selftests'

I'm not sure what's breaking ...

-- 
Kees Cook