linux-kernel - Re: linux-next: build failure after merge of the origin tree

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: Windows password security audit tool. GUI, reports in PDF.

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

Date:   Mon, 13 Sep 2021 18:29:26 -0700
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Stephen Rothwell <sfr@...b.auug.org.au>
Cc:     Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        Linux Next Mailing List <linux-next@...r.kernel.org>,
        Michael Ellerman <mpe@...erman.id.au>,
        PowerPC <linuxppc-dev@...ts.ozlabs.org>
Subject: Re: linux-next: build failure after merge of the origin tree

On Mon, Sep 13, 2021 at 5:58 PM Stephen Rothwell <sfr@...b.auug.org.au> wrote:
>
> > I have no idea why it then complains about removal of the GCC4 macros.
>
> Me neither :-(

Ooh.

So I'm looking at gcc sources, just to see if "maybe this thing is
somehow conditional".

And bingo.

In cpp_init_special_builtins(), gcc does

      if (b->value == BT_HAS_ATTRIBUTE
          && (CPP_OPTION (pfile, lang) == CLK_ASM
              || pfile->cb.has_attribute == NULL))
        continue;

which basically says that if we're pre-processing an ASM file, the
magical pre-processor symbol for __has_attribute is not defined.

I'm not sure what that 'pfile->cb.has_attribute == NULL' thing means,
but the libcpp/ChangeLog file also mentions this:

        (cpp_init_special_builtins): Don't initialize __has_attribute
        or __has_cpp_attribute if CLK_ASM or pfile->cb.has_attribute is NULL.

So this is a very very special magical thing: if building an *.S file,
__has_attribute magically goes away.

And sure enough, that's exactly what is going on. It's during that
build of arch/powerpc/boot/crt0.S, and the reason this hits on powerpc
is that in arch/powerpc/boot/Makefile we have

         -include $(srctree)/include/linux/compiler_attributes.h

as part of BOOTCFLAGS, and then it does

        BOOTAFLAGS      := -D__ASSEMBLY__ $(BOOTCFLAGS) -nostdinc

to also include that header file when building ASM files.

And our old GCC4 code silently hid this all, and made it work, because
for a *.S file  you'd then (completely illogically) get those fake
gcc-4 attribute macros.

Now, do I know *why* that ppc Makefile it does that? No. Neither do I
know why the gcc people decided to just make ASM preprocessor so
special.

But at least I understand how the odd error happens.

This was too damn subtle. When you have to go read the compiler
sources to figure things like this out, you know you are too deep.

The fix should be pretty simple: remove almost all of BOOTCFLAGS from
BOOTAFLAGS.

But sadly, "almost all" isn't "all". There's the include path stuff,
there's the ABI and endianness, and there's the bit size ones.

So I think the fix is either

 (a) remove that

         -include $(srctree)/include/linux/compiler_attributes.h

     thing entirely, and add it as required to the C files.

OR

 (b) something like this ENTIRELY UNTESTED ATTACHED patch

I will leave it to the powerpc people to make the right choice.

               Linus

View attachment "patch.diff" of type "text/x-patch" (1611 bytes)