[PATCH] arm: kirkwood: fix freeze on boot

[Tom Rini]

On Mon, Jun 23, 2025 at 10:08:16AM +0200, Jerome Forissier wrote:

> Commit 6fe50e395080 ("arm: asm/system.h: mrc and mcr need .arm if
> __thumb2__ is not set") is not a proper fix for the LTO link error
> mentioned in its description. It causes 32-bit arm instructions to be
> mixed with thumb instructions, which the Kirkwood SoCs do not support.
> For example, board_init_r() is mostly generated in Thumb-1 mode as
> expected since the build flags contain -mthumb -mthumb-interwork. The
> MCR instruction corresponding to writefr_extra_feature_reg() is also
> correcly emitted as a 32-bit ARM instruction (it cannot be encoded in
> Thumb-1 anyways). The problem is, the compiler inlines the MCR without
> generating the BX or BLX instruction which are needed to transition
> between the ARM and the Thumb-1 states. From the objdump output:
> 
> 006186a0 <board_init_r>:
> board_init_r():
> /home/jerome/work/u-boot/common/board_r.c:799
>   6186a0:       b5f0            push    {r4, r5, r6, r7, lr}
>   6186a2:       b0ab            sub     sp, #172        @ 0xac
> get_gd():
> /home/jerome/work/u-boot/./arch/arm/include/asm/global_data.h:127
>   6186a4:       464a            mov     r2, r9
> ...
> /home/jerome/work/u-boot/arch/arm/mach-kirkwood/cpu.c:242
>   619aae:       9b15            ldr     r3, [sp, #84]   @ 0x54
> writefr_extra_feature_reg():
> /home/jerome/work/u-boot/./arch/arm/include/asm/arch/cpu.h:100
>   619ab0:       ee2f3f11        mcr     15, 1, r3, cr15, cr1, {0}
>                 ^^^^^^^^
>                 32-bit ARM instruction
> 
> Further investigation is needed to understand how to fix the issue so
> that the code size is minimal for all boards. In the mean time, this
> fix disables LTO for the two problematic files (common/board_f.c and
> common/board_r.c). This makes the Kirkwood-based boards bootable again.
> The binary size is increased by 1048 bytes which is perfectly
> acceptable.
> 
> Fixes: 6fe50e395080 ("arm: asm/system.h: mrc and mcr need .arm if __thumb2__ is not set")
> Reported-by: Tony Dinh <mibodhi at gmail.com>
> Signed-off-by: Jerome Forissier <jerome.forissier at linaro.org>
> ---
> 
>  common/Makefile | 2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/common/Makefile b/common/Makefile
> index 35991562a12..f0b7eadf7d2 100644
> --- a/common/Makefile
> +++ b/common/Makefile
> @@ -19,6 +19,8 @@ obj-y += version.o
>  # # boards
>  obj-y += board_f.o
>  obj-y += board_r.o
> +CFLAGS_REMOVE_board_f.o := $(LTO_CFLAGS)
> +CFLAGS_REMOVE_board_r.o := $(LTO_CFLAGS)
>  obj-$(CONFIG_DISPLAY_BOARDINFO) += board_info.o
>  obj-$(CONFIG_DISPLAY_BOARDINFO_LATE) += board_info.o
>  

I'm sizing checking this now, but we might need a v2 with something
like:
ifeq ($(CONFIG_SYS_THUMB_BUILD),y)
... remove ...
endif

So that we don't grow *everyone* with LTO by 1KiB. I'm checking now how
bad that will go.

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20250623/2b89c9d7/attachment.sig>