Hello, Iwamatsu-san
(2012/07/06 9:06), Nobuhiro Iwamatsu wrote:
Latest rmobile code was tested by using old gcc (gcc-4.4). When we use gcc-4.6 (or later), the build is made, but does not work. This solves a problem not to work by add -march=armv5 to compiple option when we built in gcc-4.6 (or later). I tested by linaro's compiler version 2012.04-20120426.
Yes. I know recent gcc causes problem because of unaligned access. There are 2 different proposals, but neither accepted.
[PATCH] arm: enable unaligned access on ARMv7 http://lists.denx.de/pipermail/u-boot/2012-June/125754.html
[PATCH] arm: armv7: add compile option -mno-unaligned-access if available http://lists.denx.de/pipermail/u-boot/2012-July/127260.html
This patch is the third one.
--- a/arch/arm/cpu/armv7/rmobile/lowlevel_init.S +++ b/arch/arm/cpu/armv7/rmobile/lowlevel_init.S @@ -49,8 +49,6 @@ wait_interrupt: str r2, [r1]
wait_loop:
- wfi
- ldr r2, [r1, #0xC] str r2, [r1, #0x10]
I know "wfi" instruction causes compile error in -march=armv5. Without this, it works but not so good because 2nd CPU wastes cycles.
I wonder, is there any cheat such as ".word 0x...." (putting instruction code directly) or something.