Re: [PATCH] ftrace,kcfi: Separate ftrace_stub() and ftrace_stub_graph()

[Kees Cook <kees.cook@linux.dev>]

On October 18, 2022 6:18:46 AM PDT, Peter Zijlstra <peterz@infradead.org> wrote:
>On Tue, Oct 18, 2022 at 02:35:16PM +0200, Peter Zijlstra wrote:
>> --- a/include/asm-generic/vmlinux.lds.h
>> +++ b/include/asm-generic/vmlinux.lds.h
>> @@ -162,6 +162,16 @@
>>  #define PATCHABLE_DISCARDS	*(__patchable_function_entries)
>>  #endif
>>  
>> +#ifndef ARCH_SUPPORTS_CFI_CLANG
>
>#ifndef CONFIG_ARCH_..
>
>works much better as we found.

I nearly did this with an IS_ENABLED() recently. Saved by checkpatch! I wonder if it has similar checks for #ifdefs...

>
>> +/*
>> + * Simply points to ftrace_stub, but with the proper protocol.
>> + * Defined by the linker script in linux/vmlinux.lds.h
>> + */
>> +#define	FTRACE_STUB_HACK	ftrace_stub_graph = ftrace_stub;
>> +#else
>> +#define FTRACE_STUB_HACK
>> +#endif
>
>Fixed up version available at:
>
>  git://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git x86/urgent

Thanks for solving this! Just for future archeology, can you include the splat (I assume you hit a CFI splat) in the commit log, and/or how you triggered the problem? I usually find it helpful in trying to fix similar issues later, etc.

-- 
Kees Cook