Re: [tip: objtool/core] x86/insn: Support big endian cross-compiles

[Ian Rogers]

On Wed, Oct 14, 2020 at 12:29 AM Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
>
> On Mon, 12 Oct 2020 10:39:49 -0500
> Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
>
> > On Mon, Oct 12, 2020 at 09:12:36AM +0900, Masami Hiramatsu wrote:
> > > On Sat, 10 Oct 2020 12:44:15 -0500
> > > Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
> > >
> > > > On Fri, Oct 09, 2020 at 10:49:21PM +0200, Borislav Petkov wrote:
> > > > > On Fri, Oct 09, 2020 at 10:38:22PM +0200, Peter Zijlstra wrote:
> > > > > > On Wed, Oct 07, 2020 at 04:20:19PM -0000, tip-bot2 for Martin Schwidefsky wrote:
> > > > > > > The following commit has been merged into the objtool/core branch of tip:
> > > > > > >
> > > > > > > Commit-ID:     2a522b53c47051d3bf98748418f4f8e5f20d2c04
> > > > > > > Gitweb:        https://git.kernel.org/tip/2a522b53c47051d3bf98748418f4f8e5f20d2c04
> > > > > > > Author:        Martin Schwidefsky <schwidefsky@xxxxxxxxxx>
> > > > > > > AuthorDate:    Mon, 05 Oct 2020 17:50:31 +02:00
> > > > > > > Committer:     Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
> > > > > > > CommitterDate: Tue, 06 Oct 2020 09:32:29 -05:00
> > > > > > >
> > > > > > > x86/insn: Support big endian cross-compiles
> > > > > > >
> > > > > > > x86 instruction decoder code is shared across the kernel source and the
> > > > > > > tools. Currently objtool seems to be the only tool from build tools needed
> > > > > > > which breaks x86 cross compilation on big endian systems. Make the x86
> > > > > > > instruction decoder build host endianness agnostic to support x86 cross
> > > > > > > compilation and enable objtool to implement endianness awareness for
> > > > > > > big endian architectures support.
> > > > > > >
> > > > > > > Signed-off-by: Martin Schwidefsky <schwidefsky@xxxxxxxxxx>
> > > > > > > Co-developed-by: Vasily Gorbik <gor@xxxxxxxxxxxxx>
> > > > > > > Signed-off-by: Vasily Gorbik <gor@xxxxxxxxxxxxx>
> > > > > > > Acked-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
> > > > > > > Signed-off-by: Josh Poimboeuf <jpoimboe@xxxxxxxxxx>
> > > > > >
> > > > > > This commit breaks the x86 build with CONFIG_X86_DECODER_SELFTEST=y.
> > > > > >
> > > > > > I've asked Boris to truncate tip/objtool/core.
> > > > >
> > > > > Yeah, top 4 are gone until this is resolved.
> > > >
> > > > Masami, I wonder if we even need these selftests anymore?  Objtool
> > > > already decodes the entire kernel.
> > >
> > > No, they have different roles. The selftest checks if the decoder
> > > works correctly by comparing with the output of objdump.
> > >
> > > As far as I can see, the objtool relies on the sanity of the decoder
> > > (it trusts the output of the decoder).
> >
> > Ok.  I wonder if we should move the decoder selftest to the 'tools'
> > subdirectory.
>
> It is in the arch/x86/tools, so it is already in a kind of tools :)
> But yeah, it was considered to be used only on x86. But if someone
> start trying to run it on non-x86, cross compiling, we need to
> reconsider that.
>
> Thank you,
>
> --
> Masami Hiramatsu <mhiramat@xxxxxxxxxx>

There is undefined behavior that is present in the x86 insn.c code as
described in:
https://lore.kernel.org/lkml/20190724184512.162887-4-nums@xxxxxxxxxx/

I resent this patch fixing other potential undefined behavior:
https://lore.kernel.org/lkml/20201015062148.1437894-1-irogers@xxxxxxxxxx/T/#t

The misaligned loads will likely break on anything non-x86.

Thanks,
Ian