Re: PROBLEM: 5.11.0-rc7 fails to compile with error: ‘-mindirect-branch’ and ‘-fcf-protection’ are not compatible

On Mon, 8 Feb 2021 at 01:40, Stuart Little [off-list ref] wrote:

And for good measure: reverting that commit

20bf2b378729c4a0366a53e2018a0b70ace94bcd

flagged by the bisect right on top of the current tree compiles fine.

On Sun, Feb 07, 2021 at 07:26:01PM -0500, Stuart Little wrote:

quoted

The result of the bisect on the issue reported in the previous message:

--- cut ---

20bf2b378729c4a0366a53e2018a0b70ace94bcd is the first bad commit
commit 20bf2b378729c4a0366a53e2018a0b70ace94bcd
Author: Josh Poimboeuf [off-list ref]
Date:   Thu Jan 28 15:52:19 2021 -0600

    x86/build: Disable CET instrumentation in the kernel

    With retpolines disabled, some configurations of GCC, and specifically
    the GCC versions 9 and 10 in Ubuntu will add Intel CET instrumentation
    to the kernel by default. That breaks certain tracing scenarios by
    adding a superfluous ENDBR64 instruction before the fentry call, for
    functions which can be called indirectly.

    CET instrumentation isn't currently necessary in the kernel, as CET is
    only supported in user space. Disable it unconditionally and move it
    into the x86's Makefile as CET/CFI... enablement should be a per-arch
    decision anyway.

     [ bp: Massage and extend commit message. ]

    Fixes: 29be86d7f9cb ("kbuild: add -fcf-protection=none when using retpoline flags")
    Reported-by: Nikolay Borisov [off-list ref]
    Signed-off-by: Josh Poimboeuf [off-list ref]
    Signed-off-by: Borislav Petkov [off-list ref]
    Reviewed-by: Nikolay Borisov [off-list ref]
    Tested-by: Nikolay Borisov [off-list ref]
    Cc: [off-list ref]
    Cc: Seth Forshee [off-list ref]
    Cc: Masahiro Yamada [off-list ref]
    Link: https://lkml.kernel.org/r/20210128215219.6kct3h2eiustncws@treble

 Makefile          | 6 ------
 arch/x86/Makefile | 3 +++
 2 files changed, 3 insertions(+), 6 deletions(-)

--- end ---

On Sun, Feb 07, 2021 at 06:31:22PM -0500, Stuart Little wrote:

quoted

I am trying to compile on an x86_64 host for a 32-bit system; my config is at

https://termbin.com/v8jl

I am getting numerous errors of the form

./include/linux/kasan-checks.h:17:1: error: ‘-mindirect-branch’ and ‘-fcf-protection’ are not compatible

This is an empty static inline function...

quoted

quoted

and

./include/linux/kcsan-checks.h:143:6: error: ‘-mindirect-branch’ and ‘-fcf-protection’ are not compatible

... and so is this. I think these have very little to do with the
problem that you reported. My guess is they show up because these are
included very early.

quoted

quoted

and

./arch/x86/include/asm/arch_hweight.h:16:1: error: ‘-mindirect-branch’ and ‘-fcf-protection’ are not compatible

(those include files indicated whom I should add to this list; apologies if this reaches you in error).

The full log of the build is at

https://termbin.com/wbgs

The commonality between all these errors is that they originate from
compiling arch/x86/entry/vdso/vdso32/vclock_gettime.c.

Is the build system adding special flags for vdso? In which case, it's
probably just GCC complaining about every function definition (static
inline or otherwise) for that TU if (for whatever reason) it's
delaying the flag compatibility check until it inspects function
attributes.

And indeed, I can see:

  RETPOLINE_VDSO_CFLAGS_GCC := -mindirect-branch=thunk-inline
-mindirect-branch-register

And taking any test source with even an empty function definition:

  > gcc -mindirect-branch=thunk-inline -fcf-protection test.c
  > test.c: In function ‘main’:
  > test.c:6:1: error: ‘-mindirect-branch’ and ‘-fcf-protection’ are
not compatible

quoted

quoted

5.11.0-rc6 built fine last week on this same setup.

Thanks,
-- Marco