Decoding BLX instruction on ARM/Thumb (IOS)

Question

I've read through much of the ARM documentation but still having trouble decoding a BLX instruction. Here are two examples:

__text:0000347C 02 F0 B2 ED                 BLX             _objc_msgSend

__text:0000469C 01 F0 A2 EC                 BLX             _objc_msgSend

Both of these are supposed to go to the same place, virtual address 0x5FE4 as seen here:

__symbolstub1:00005FE4 38 F0 9F E5                 LDR             PC, =__imp__objc_msgSend

However, I can't figure out what calculation is used to get from the above two addresses (0x347C and 0x469C) using their instruction bytes. According to the ARM documentation its supposed to be a relative jump, using a right shift by 2, but the numbers don't work out.

Can anyone help out?

Answer 1

First, the instruction is being printed as two little-endian 16-bit fields. To match the byte ordering in the ARM reference manual, you need to byteswap each of the fields. For the first instruction, that gives:

F0 02 ED B2

or

11110000000000101110110110110010.

This is encoding T2 of the BLX instruction. Breaking this into the fields identified in the ARM:

11110  0  0000000010  11  1  0  1  1011011001  0  
       S    imm10H        J1    J2   imm10L

Then follow the instructions for interpretation of the fields:

I1 = NOT(J1 EOR S) = 0
I2 = NOT(J2 EOR S) = 0

imm32 = SignExtend(S:I1:I2:imm10H:imm10L:00)
      = SignExtend(0000000000010101101100100)
      = 0x00002b64

Which is precisely 0x5FE4 - 0x3480 (remember, the PC is 4 bytes ahead in Thumb/Thumb 2).

I trust you can work through the second example yourself.