Question about unaligned memory access on ARMv7

Hi,

My linux kernel complained about the user space binaries doing unaligned memory access.

Here is one of the log:

[ 1056.522396] Alignment trap: rsyslogd (1400) PC=0x4004450e

Instr=0xe8541f00 Address=0x409f84f5 FSR 0x001

$ echo "0xe8 0x54 0x1f 0x00"|/opt/local/ibos/bin/llvm-mc --disassemble
-triple=thumb-apple-darwin9
     strb    r0, [r5, r3]
     movs    r7, r3

I'm using the NVIDIA Harmony board, which has a Cortex-A9MP dual core CPU. According to the specification, this instruction is legitimate and shouldn't fault.

I checked in arch/arm/mm/alignment.c, the function do_alignment_t32_to_handler will ignore this because the CPU should be able to handle it.

I also checked with the newest version of arch/arm/mm/alignment and I noticed no changes.

Am I missing some erratas that can possibly fix this bug?

PS: I'm using a sufficiently old (2.6.32.9) kernel and I'm unable to update it because a bug in the bootloader. :-(

Any suggestions are appreciated.

Cheers,
Haohui

That is an LDREX instruction. It does not support unaligned addresses.

You got the byte order wrong there.

Fix your application.

M?ns Rullg?rd
mans at mansr.com