Re: [PATCH 1/3] retpolines: Only enable retpoline when compiler support it

[Zhenzhong Duan <zhenzhong.duan@oracle.com>]

On 2018/10/30 16:32, Peter Zijlstra wrote:
> On Mon, Oct 29, 2018 at 11:55:04PM -0700, Zhenzhong Duan wrote:
>> Since retpoline capable compilers are widely available, make
>> CONFIG_RETPOLINE hard depend on it.
>>
>> Change KBUILD to use CONFIG_RETPOLINE_SUPPORT to avoid conflict with
>> CONFIG_RETPOLINE which is used by kernel.
>>
>> With all that stuff, the check of RETPOLINE is changed to
>> CONFIG_RETPOLINE.
> 
> So what happens when we select CONFIG_RETPOLINE but do not have
> RETPOLINE_SUPPORT ? From a quick reading we'll silently build a
> !retpoline kernel. I would expect a build failure.

CONFIG_RETPOLINE is only defined when CONFIG_RETPOLINE_SUPPORT is 
selected. See below chunk.

--- a/arch/x86/Makefile
+++ b/arch/x86/Makefile
@@ -221,9 +221,10 @@ KBUILD_CFLAGS += -Wno-sign-compare
  KBUILD_CFLAGS += -fno-asynchronous-unwind-tables

  # Avoid indirect branches in kernel to deal with Spectre
-ifdef CONFIG_RETPOLINE
+ifdef CONFIG_RETPOLINE_SUPPORT
  ifneq ($(RETPOLINE_CFLAGS),)
-  KBUILD_CFLAGS += $(RETPOLINE_CFLAGS) -DRETPOLINE
+  KBUILD_CFLAGS += $(RETPOLINE_CFLAGS) -DCONFIG_RETPOLINE
+  KBUILD_AFLAGS += -DCONFIG_RETPOLINE
  endif
  endif


Thanks
Zhenzhong