Re: [PATCH v3] kbuild: Add support to generate LLVM bitcode files

From: Masahiro Yamada <yamada.masahiro@socionext.com>
To: Matthias Kaehlcke <mka@chromium.org>
Cc: "Michal Marek" <mmarek@suse.com>,
	"Emese Revfy" <re.emese@gmail.com>,
	"Kees Cook" <keescook@chromium.org>,
	"Behan Webster" <behanw@converseincode.com>,
	"Luis R . Rodriguez" <mcgrof@kernel.org>,
	"Vinícius Tinti" <viniciustinti@gmail.com>,
	"Kyeongmin Cho" <korea.drzix@gmail.com>,
	"Linux Kernel Mailing List" <linux-kernel@vger.kernel.org>,
	"Linux Kbuild mailing list" <linux-kbuild@vger.kernel.org>,
	"Grant Grundler" <grundler@chromium.org>,
	"Michael Davidson" <md@google.com>,
	"Greg Hackmann" <ghackmann@google.com>,
	"Peter Foley" <pefoley2@pefoley.com>
Subject: Re: [PATCH v3] kbuild: Add support to generate LLVM bitcode files
Date: Sun, 23 Apr 2017 15:57:49 +0900	[thread overview]
Message-ID: <CAK7LNAT-tDQGa4b2jugMDb8umOLLoe0GE_v6n5AuURmLG_TL2Q@mail.gmail.com> (raw)
In-Reply-To: <20170421195502.GL128305@google.com>

Hi Matthias,

2017-04-22 4:55 GMT+09:00 Matthias Kaehlcke <mka@chromium.org>:
> Hi Masahiro,
>
> El Fri, Apr 21, 2017 at 02:02:46PM +0900 Masahiro Yamada ha dit:
>
>> 2017-04-05 2:27 GMT+09:00 Matthias Kaehlcke <mka@chromium.org>:
>> > From: Vinícius Tinti <viniciustinti@gmail.com>
>> >
>> > Add rules to kbuild in order to generate LLVM bitcode files with the .ll
>> > extension when using clang.
>>
>>
>> First, I'd like to be sure about the terminology "LLVM bitcode"
>> because "bitcode" sounds like human-unreadable binary.
>>
>>
>> For example, 'man llvm-as' says:
>>     llvm-as  is  the  LLVM  assembler.  It reads a file containing
>>     human-readable LLVM assembly language, translates it to LLVM
>>     bitcode, and writes the result into a file or to standard output.
>>
>>
>> As far as I understood:
>>
>> *.ll   -  LLVM assembly  (human readable file)
>> *.bc   -  LLVM bitcode   (binary file)
>>
>> Is this correct?
>
> Yes, the terminology should be changed to talk about 'LLVM assembly'.
>
>> >   # from c code
>> >   CC=clang make kernel/pid.ll
>>
>> This does not work because CC is overridden in the top-level Makefile.
>> It should be
>>     make CC=clang kernel/pid.ll
>
> Will change
>
>> >   # from asm code
>> >   CC=clang make arch/x86/kernel/preempt.ll
>>
>> arch/x86/kernel/preempt.* does not exist
>> (at least in the latest tree).
>>
>>
>>
>>
>> > +
>> > +quiet_cmd_as_ll_S = CPP $(quiet_modtag) $@
>> > +      cmd_as_ll_S = $(CPP) $(a_flags)   -o $@ $<
>> > +
>> > +$(obj)/%.ll: $(src)/%.S FORCE
>> > +       $(call if_changed_dep,as_ll_S)
>> > +
>>
>> I could not understand how this rule can convert
>> architecture-specific assembly to LLVM intermediate expression.
>>
>> This is just pre-processing *.S file.
>>
>>
>> Actually, this is completely the same as the rule *.S -> *.s
>>
>> quiet_cmd_cpp_s_S = CPP $(quiet_modtag) $@
>> cmd_cpp_s_S       = $(CPP) $(a_flags) -o $@ $<
>>
>> $(obj)/%.s: $(src)/%.S FORCE
>>       $(call if_changed_dep,cpp_s_S)
>
> Indeed, unsurprisingly the content of a .ll file generated from a .S
> is the same as the corresponding .s.
>
> Besides the Makefile rules it isn't clear to me how assembly would be
> converted to LLVM IR. I suggest to remove the rules for assembly.
>

I agree.  This rule should be removed.

-- 
Best Regards
Masahiro Yamada