Re: [PATCH] riscv: boot: Always make Image from vmlinux, not vmlinux.unstripped

From: Nathan Chancellor <nathan@kernel.org>
To: Vivian Wang <wangruikang@iscas.ac.cn>
Cc: Paul Walmsley <pjw@kernel.org>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	Albert Ou <aou@eecs.berkeley.edu>,
	Alexandre Ghiti <alex@ghiti.fr>,
	Alexey Gladkov <legion@kernel.org>,
	Masahiro Yamada <masahiroy@kernel.org>,
	Nicolas Schier <nsc@kernel.org>, Han Gao <gaohan@iscas.ac.cn>,
	linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
	linux-kbuild@vger.kernel.org, stable@vger.kernel.org
Subject: Re: [PATCH] riscv: boot: Always make Image from vmlinux, not vmlinux.unstripped
Date: Tue, 30 Dec 2025 13:03:36 -0700	[thread overview]
Message-ID: <20251230200336.GA4062669@ax162> (raw)
In-Reply-To: <20251230-riscv-vmlinux-not-unstripped-v1-1-15f49df880df@iscas.ac.cn>

On Tue, Dec 30, 2025 at 09:39:17PM +0800, Vivian Wang wrote:
> Since commit 4b47a3aefb29 ("kbuild: Restore pattern to avoid stripping
> .rela.dyn from vmlinux") vmlinux has .rel*.dyn preserved. Therefore, use
> vmlinux to produce Image, not vmlinux.unstripped.
> 
> Doing so fixes booting a RELOCATABLE=y Image with kexec. The problem is
> caused by this chain of events:
> 
> - Since commit 3e86e4d74c04 ("kbuild: keep .modinfo section in
>   vmlinux.unstripped"), vmlinux.unstripped gets a .modinfo section.
> - The .modinfo section has SHF_ALLOC, so it ends up in Image, at the end
>   of it.
> - The Image header's image_size field does not expect to include
>   .modinfo and does not account for it, since it should not be in Image.
> - If .modinfo is large enough, the file size of Image ends up larger
>   than image_size, which eventually leads to it failing
>   sanity_check_segment_list().
> 
> Using vmlinux instead of vmlinux.unstripped means that the unexpected
> .modinfo section is gone from Image, fixing the file size problem.
> 
> Cc: stable@vger.kernel.org
> Fixes: 3e86e4d74c04 ("kbuild: keep .modinfo section in vmlinux.unstripped")
> Signed-off-by: Vivian Wang <wangruikang@iscas.ac.cn>

Yes, this seems reasonable. I suspect this could have been done after
commit e9d86b8e17e7 ("scripts: Do not strip .rela.dyn section") in 6.15
since it has the same effect that 4b47a3aefb29 does but there was no
visible problem.

Reviewed-by: Nathan Chancellor <nathan@kernel.org>

> ---
>  arch/riscv/boot/Makefile | 4 ----
>  1 file changed, 4 deletions(-)
> 
> diff --git a/arch/riscv/boot/Makefile b/arch/riscv/boot/Makefile
> index bfc3d0b75b9b..5301adf5f3f5 100644
> --- a/arch/riscv/boot/Makefile
> +++ b/arch/riscv/boot/Makefile
> @@ -31,11 +31,7 @@ $(obj)/xipImage: vmlinux FORCE
>  
>  endif
>  
> -ifdef CONFIG_RELOCATABLE
> -$(obj)/Image: vmlinux.unstripped FORCE
> -else
>  $(obj)/Image: vmlinux FORCE
> -endif
>  	$(call if_changed,objcopy)
>  
>  $(obj)/Image.gz: $(obj)/Image FORCE
> 
> ---
> base-commit: 8f0b4cce4481fb22653697cced8d0d04027cb1e8
> change-id: 20251230-riscv-vmlinux-not-unstripped-30ec0c930fd2
> 
> Best regards,
> -- 
> Vivian "dramforever" Wang
> 

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv