llvm dev - Oct 2019 - Is it possible to generate a whole kernel bitcode for Linux using clang?

Hello,

With clang/llvm 9.0 adding support for ASM GOTOs, it is now possible to
compile the latest Linux kernel with clang without any modification.
However, is it possible to generate a single bitcode for the whole Linux
kernel? Earlier approaches would involve hacks, like compiling separate
bitcode files and then linking with llvm-link, or using wllvm, or using
some other wrapper. Frequently, the resulting bitcode was not functional
due to different parts being missing. Did that situation change at all? Can
we generate a function, monolithic, self-contained bitcode using something
like libLTO? Any help would be appreciated.

-- 

JT
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20191010/165642d9/attachment.html>

Hi Jason,

On Thu, 10 Oct 2019 at 10:39, Jason Thomas via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
> Frequently, the resulting bitcode was not functional due to different parts
being missing. Did that situation change at all? Can we generate a function,
monolithic, self-contained bitcode using something like libLTO? Any help would
be appreciated.
You can get the linker to emit the last bitcode that exists (e.g. via
--plugin-opt save-temps or emit-llvm I think), but it still won't be
self contained for something like a kernel. There's no representation
in IR for objects that come from .s files or the linker script for a
start.

Cheers.

Tim.

Hi Tim,

Now I understand why it can't be a self-contained IR. However, just to emit
the bitcode of individual files, I tried compiling with

make CC=clang HOSTCC=clang CFLAGS=" -g -flto -std=gnu99 "
LDFLAGS=" -flto
-fuse-ld=gold -Wl,-plugin-opt=save-temps" menuconfig
make CC=clang HOSTCC=clang CFLAGS=" -g -flto -std=gnu99 "
LDFLAGS=" -flto
-fuse-ld=gold -Wl,-plugin-opt=save-temps" -j8

The compilation with clang is still going on. But, it doesn't look like it
generates the IR at all. Whatever *.o files the build is producing are
native ELF files.

On Thu, 10 Oct 2019 at 02:54, Tim Northover <t.p.northover at gmail.com>
wrote:
> Hi Jason,
>
> On Thu, 10 Oct 2019 at 10:39, Jason Thomas via llvm-dev
> <llvm-dev at lists.llvm.org> wrote:
> > Frequently, the resulting bitcode was not functional due to different
> parts being missing. Did that situation change at all? Can we generate a
> function, monolithic, self-contained bitcode using something like libLTO?
> Any help would be appreciated.
>
> You can get the linker to emit the last bitcode that exists (e.g. via
> --plugin-opt save-temps or emit-llvm I think), but it still won't be
> self contained for something like a kernel. There's no representation
> in IR for objects that come from .s files or the linker script for a
> start.
>
> Cheers.
>
> Tim.
>

-- 

JT
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20191010/5abd97ad/attachment.html>