Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Arvind Sankar <nivedita@alum.mit.edu>
Date:   Fri, 10 Apr 2020 11:16:12 -0400
To:     Ard Biesheuvel <ardb@kernel.org>
Cc:     Arvind Sankar <nivedita@alum.mit.edu>,
        Brian Gerst <brgerst@gmail.com>,
        linux-efi <linux-efi@vger.kernel.org>,
        Ingo Molnar <mingo@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Arnd Bergmann <arnd@arndb.de>, Borislav Petkov <bp@suse.de>,
        Colin Ian King <colin.king@canonical.com>,
        Gary Lin <glin@suse.com>, Jiri Slaby <jslaby@suse.cz>,
        Sergey Shatunov <me@prok.pw>, Takashi Iwai <tiwai@suse.de>
Subject: Re: [PATCH 3/9] efi/x86: Move efi stub globals from .bss to .data
Message-ID: <20200410151612.GA970420@rani.riverdale.lan>
References: <20200409130434.6736-1-ardb@kernel.org>
 <20200409130434.6736-4-ardb@kernel.org>
 <CAMzpN2gJWwVun1Kp6vGuza9LM5KpB=0EwsP8x8eOJQuDGh38Hg@mail.gmail.com>
 <CAMzpN2jFbf8k99pWaTYRBmSB+iNAKYsufjEhqO6Vv0qxAcHyGA@mail.gmail.com>
 <20200409210847.GA1312580@rani.riverdale.lan>
 <CAMj1kXFhtK=FRDKBE5OtenNEtpK=kVwyo+0nqJZ_K80RmtYxEg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <CAMj1kXFhtK=FRDKBE5OtenNEtpK=kVwyo+0nqJZ_K80RmtYxEg@mail.gmail.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Apr 10, 2020 at 10:20:42AM +0200, Ard Biesheuvel wrote:
> On Thu, 9 Apr 2020 at 23:08, Arvind Sankar <nivedita@alum.mit.edu> wrote:
> >
> > On Thu, Apr 09, 2020 at 04:53:07PM -0400, Brian Gerst wrote:
> > > > Can we use the -fno-zero-initialized-in-bss compiler flag instead of
> > > > explicitly marking global variables?
> > >
> > > Scratch that.  Apparently it only works when a variable is explicitly
> > > initialized to zero.
> > >
> > > --
> > > Brian Gerst
> >
> > Right, there doesn't seem to be a compiler option to turn off the use of
> > .bss altogether.
> 
> Yeah. I'll try to come up with a way to consolidate this a bit across
> architectures (which is a bit easier now that all of the EFI stub C
> code lives in the same place). It is probably easiest to use a section
> renaming trick similar to the one I added for ARM (as Arvind suggested
> as well, IIRC), and get rid of the per-symbol annotations altogether.

Does that work for 32-bit ARM, or does it need to be .data to tell the
compiler to avoid generating GOT references? If that's fine, we don't
actually need to rename sections -- linker script magic is enough. For
eg, the below pulls the EFI stub bss into .data for x86 without the need
for the annotations.

diff --git a/arch/x86/boot/compressed/vmlinux.lds.S b/arch/x86/boot/compressed/vmlinux.lds.S
index 508cfa6828c5..e324819c95bc 100644
--- a/arch/x86/boot/compressed/vmlinux.lds.S
+++ b/arch/x86/boot/compressed/vmlinux.lds.S
@@ -52,6 +52,7 @@ SECTIONS
 		_data = . ;
 		*(.data)
 		*(.data.*)
+		drivers/firmware/efi/libstub/lib.a:(.bss .bss.*)
 		_edata = . ;
 	}
 	. = ALIGN(L1_CACHE_BYTES);