Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210802234304.3519577-1-ndesaulniers@google.com>
 <CAK7LNATWm94K=UcSNn88PyPST38rfe_31vLLpDf=ERPnKnBvVA@mail.gmail.com> <CAKwvOdkVuzD68bQuAWS7fS0tat_wdR0bxb=Eh=Te84RtoRgZ8g@mail.gmail.com>
In-Reply-To: <CAKwvOdkVuzD68bQuAWS7fS0tat_wdR0bxb=Eh=Te84RtoRgZ8g@mail.gmail.com>
From:   Masahiro Yamada <masahiroy@kernel.org>
Date:   Fri, 6 Aug 2021 08:48:07 +0900
Message-ID: <CAK7LNATJRTD7dowBosb1-560C6-NxoorjxAYMzV5SZPd35GAtg@mail.gmail.com>
Subject: Re: [PATCH] scripts/Makefile.clang: default to LLVM_IAS=1
To:     Nick Desaulniers <ndesaulniers@google.com>
Cc:     Nathan Chancellor <nathan@kernel.org>,
        Khem Raj <raj.khem@gmail.com>,
        Michal Marek <michal.lkml@markovi.net>,
        Jonathan Corbet <corbet@lwn.net>,
        Paul Walmsley <paul.walmsley@sifive.com>,
        Palmer Dabbelt <palmer@dabbelt.com>,
        Albert Ou <aou@eecs.berkeley.edu>,
        clang-built-linux <clang-built-linux@googlegroups.com>,
        Linux Kbuild mailing list <linux-kbuild@vger.kernel.org>,
        "open list:DOCUMENTATION" <linux-doc@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        "open list:SIFIVE DRIVERS" <linux-riscv@lists.infradead.org>,
        Matthew Wilcox <willy@infradead.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Aug 6, 2021 at 3:40 AM Nick Desaulniers <ndesaulniers@google.com> wrote:
>
> On Thu, Aug 5, 2021 at 8:16 AM Masahiro Yamada <masahiroy@kernel.org> wrote:
> >
> > On Tue, Aug 3, 2021 at 8:43 AM Nick Desaulniers <ndesaulniers@google.com> wrote:
> > >
> > > diff --git a/Makefile b/Makefile
> > > index 444558e62cbc..b24b48c9ebb7 100644
> > > --- a/Makefile
> > > +++ b/Makefile
> > > @@ -845,7 +845,7 @@ else
> > >  DEBUG_CFLAGS   += -g
> > >  endif
> > >
> > > -ifneq ($(LLVM_IAS),1)
> > > +ifeq ($(LLVM_IAS),0)
> > >  KBUILD_AFLAGS  += -Wa,-gdwarf-2
> > >  endif
> > >
> > > diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile
> > > index bc74afdbf31e..807f7c94bc6f 100644
> > > --- a/arch/riscv/Makefile
> > > +++ b/arch/riscv/Makefile
> > > @@ -41,7 +41,7 @@ endif
> > >  ifeq ($(CONFIG_LD_IS_LLD),y)
> > >         KBUILD_CFLAGS += -mno-relax
> > >         KBUILD_AFLAGS += -mno-relax
> > > -ifneq ($(LLVM_IAS),1)
> > > +ifeq ($(LLVM_IAS),0)
> > >         KBUILD_CFLAGS += -Wa,-mno-relax
> > >         KBUILD_AFLAGS += -Wa,-mno-relax
> > >  endif
> >
> >
> >
> > Please drop these two hunks.
> >
> > I will apply my patch instead.
> > https://lore.kernel.org/patchwork/patch/1472580/
>
> Sure.  Will send a v2 with Matthew's suggestion as well.
>
> > When we negate a flag that is enabled by default,
> > which is a common way?
> >  - set it to '0'
> >  - set is to empty
> >
> >
> > So, I was wondering if we should support
> > not only LLVM_IAS=0 but also LLVM_IAS=.
> >
> > What do you think?
>
> LLVM_IAS= looks weird (so I agree with Khem's response), but if it's
> common/expected then maybe if there's a way to write a concise check
> for either =<blank> or =0?  I don't feel strongly about how it's
> specified to disable the integrated assembler, but let's sort that out
> _before_ I send a v2.
>
> How can you tell the difference between `make CC=clang` and `make
> CC=clang LLVM_IAS=`? (maybe that doesn't matter here, as either imply
> "don't use clang as the assembler when compiling with clang")


$(origin LLVM_IAS) knows the difference.

make CC=clang            ->  $(origin LLVM_IAS) is 'undefined'
make CC=clang LLVM_IAS=  ->  $(origin LLVM_IAS) is 'command line'
LLVM_IAS= make CC=clang  ->  $(origin LLVM_IAS) is 'environment'


The following patch makes both LLVM_IAS= and LLVM_IAS=0
work for disabling the integrated assembler.


@@ -22,6 +22,9 @@ else
 CLANG_FLAGS    += --target=$(notdir $(CROSS_COMPILE:%-=%))
 endif # CROSS_COMPILE

+# The integrated assembler is enabled by default.
+LLVM_IAS ?= 1
+
 ifeq ($(LLVM_IAS),1)
 CLANG_FLAGS    += -integrated-as
 else


But, I am not pretty sure if it is a good idea...

Perhaps, it is better to require LLVM_IAS=0
as Khem mentioned.


Another way for avoiding ambiguity is, perhaps
LLVM_IAS_DISABLE=1 instead of LLVM_IAS=0.


-- 
Best Regards
Masahiro Yamada