Received-SPF: pass (google.com: domain of linux-kernel+bounces-153900-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Date: Mon, 22 Apr 2024 15:13:59 -0400
From: Charlie Jenkins <charlie@rivosinc.com>
To: Zong Li <zong.li@sifive.com>
Cc: paul.walmsley@sifive.com, palmer@dabbelt.com, aou@eecs.berkeley.edu,
	schwab@suse.de, linux-riscv@lists.infradead.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH] Revert "riscv: disable generation of unwind tables"
Message-ID: <Zia29wYvScRVdXUP@ghost>
References: <20240416013138.28760-1-zong.li@sifive.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20240416013138.28760-1-zong.li@sifive.com>

On Tue, Apr 16, 2024 at 09:31:38AM +0800, Zong Li wrote:
> This reverts commit 2f394c0e7d1129a35156e492bc8f445fb20f43ac.
> 
> RISC-V has supported the complete relocation types in module loader by
> '8fd6c5142395 ("riscv: Add remaining module relocations")'.
> Now RISC-V port can enable unwind tables in case eh_frame parsing is
> needed.
> 
> Signed-off-by: Zong Li <zong.li@sifive.com>
> ---
>  arch/riscv/Makefile | 3 ---
>  1 file changed, 3 deletions(-)
> 
> diff --git a/arch/riscv/Makefile b/arch/riscv/Makefile
> index 5b3115a19852..9216bf8a2691 100644
> --- a/arch/riscv/Makefile
> +++ b/arch/riscv/Makefile
> @@ -94,9 +94,6 @@ ifeq ($(CONFIG_CMODEL_MEDANY),y)
>  	KBUILD_CFLAGS += -mcmodel=medany
>  endif
>  
> -# Avoid generating .eh_frame sections.
> -KBUILD_CFLAGS += -fno-asynchronous-unwind-tables -fno-unwind-tables
> -

There are a lot of orphaned sections created by this, which throws a lot
of warnings. These warnings will only be present on GCC-13+ since
-fno-asynchronous-unwind-tables was a default before GCC-13.

The .eh_frame orphaned sections can be resolved by including

eh_frame : {*(.eh_frame)}

in arch/riscv/kernel/vmlinux.lds.S. arm64 places it in the init section,
but outside init_data and init_text, so right after the .alternative
definition seems like a good place.

init.eh_frame sections in drivers/firmware/efi/libstub also complain
about being orphaned, and other architectures have disabled unwind
tables for that compilation unit, so -fno-unwind-tables and
-fno-asynchronous-unwind-tables should be added to
drivers/firmware/efi/libstub/Makefile for riscv.

This also causes the resulting image (with defconfig) to increase from
22M to 24M. There are currently no users of unwind tables in riscv so
this should be hidden behind a config. arm64 uses this config as:

# Avoid generating .eh_frame* sections.
ifneq ($(CONFIG_UNWIND_TABLES),y)
KBUILD_CFLAGS   += -fno-asynchronous-unwind-tables -fno-unwind-tables
KBUILD_AFLAGS   += -fno-asynchronous-unwind-tables -fno-unwind-tables
else
KBUILD_CFLAGS   += -fasynchronous-unwind-tables
KBUILD_AFLAGS   += -fasynchronous-unwind-tables
endif

Using this same config name for riscv would allow riscv to use the
standard .eh_frame code introduced in
https://lore.kernel.org/linux-arm-kernel/20221027155908.1940624-2-ardb@kernel.org/.

This config is only enabled if some other config selects it in arm64,
but riscv does not contain any users so maybe this config can be left
unselected until there is a user of unwind tables in riscv.

- Charlie

>  # The RISC-V attributes frequently cause compatibility issues and provide no
>  # information, so just turn them off.
>  KBUILD_CFLAGS += $(call cc-option,-mno-riscv-attribute)
> -- 
> 2.17.1
> 
> 
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv