AVR Disassembler for the 8-bit AVRs.
Generates AVRASM assembly source out of the input file provided. Alternatively, it generates the listing with the word addresses and instruction words along with the assembly source. Output is written to stdout. In case of a successful run, the exit status is 0. Otherwise it's 1. Error messages gets written to stderr.
- Maximum portability. Written in C and uses libc only, no extra libraries needed.
- Loosely coupled modular design, with clearly separated parts makes it easy to extend with new input format parsers without touching other parts.
You have an unlocked 8-bit AVR microcontroller, and you want to tinker with the code but you don't have the source.
- Use
avrdudeto extract the flash contents into an ihex file. - Use
avrdisto disassemble the code. - Use a code editor or IDE to edit the asm file.
- Use
avrasm2.exeoravraassemblers to compile the modified source code into an ihex file. - Use
avrdudeto flash the modified firmware.
First make the project.
$ cd avrdis
$ make
$ make install
This will install the avrdis executable into /usr/local/bin by default.
In case you wish to install it elsewhere, you can set a different PREFIX.
$ make install PREFIX=~/temp
This will install the avrdis executable into ~/temp/bin.
You can use the -h option to show the usage.
$ avrdis -h
AVR Disassembler for the 8-bit AVRs. v1.0.0 (c) Imre Horvath, 2024
Usage: avrdis [options] inputfile
Currently supported inputfile types are: IHEX
IHEX: Intel hex format, file should have an extension .hex
Options:
-h : Show this usage info and exit.
-l : List disabled regions, word addresses and raw instructions together with the disassembled code.
-e nnnn:nnnn : Enable disassembly of otherwise disabled region. Multiple options are possible.
Use hex numbers. For reference, see listing of disabled regions in listing mode.
Consider an example firmware foo.hex with the following content.
$ cat foo.hex
:020000020000FC
:0200000004C03A
:0200040018954D
:10000800189503B103701127EDE0F0E0E00FF11F40
:10001800099403C003C003C003C003C0F2CFF1CFEB
:10002800F0CFF0E0E0E4D0E0C0E60AE0C89509923D
:1000380031960A95D9F7E5CF0001020304050607B2
:020048000809A5
:00000001FF
Run avrdis to dissassemble it.
$ avrdis foo.hex
.org 0x0000
rjmp L0
.org 0x0002
reti
.org 0x0004
.dw 0x9518
L0: in r16, 0x03
andi r16, 3
clr r17
ldi r30, 13
ldi r31, 0
add r30, r16
adc r31, r17
ijmp
.dw 0xc003
.dw 0xc003
.dw 0xc003
.dw 0xc003
.dw 0xc003
.dw 0xcff2
.dw 0xcff1
.dw 0xcff0
.dw 0xe0f0
.dw 0xe4e0
.dw 0xe0d0
.dw 0xe6c0
.dw 0xe00a
.dw 0x95c8
.dw 0x9209
.dw 0x9631
.dw 0x950a
.dw 0xf7d9
.dw 0xcfe5
.dw 0x0100
.dw 0x0302
.dw 0x0504
.dw 0x0706
.dw 0x0908
Notice that there are a lot of not disassembled instruction words with .dw 0xnnnn after the ijmp instruction in the output! To get more insight, use the -l option.
$ avrdis -l foo.hex
0x0004:0x0004
0x000d:0x0024
C:00000 c004 rjmp L0
C:00002 9518 reti
C:00004 9518 .dw 0x9518
C:00005 b103 L0: in r16, 0x03
C:00006 7003 andi r16, 3
C:00007 2711 clr r17
C:00008 e0ed ldi r30, 13
C:00009 e0f0 ldi r31, 0
C:0000a 0fe0 add r30, r16
C:0000b 1ff1 adc r31, r17
C:0000c 9409 ijmp
C:0000d c003 .dw 0xc003
C:0000e c003 .dw 0xc003
C:0000f c003 .dw 0xc003
C:00010 c003 .dw 0xc003
C:00011 c003 .dw 0xc003
C:00012 cff2 .dw 0xcff2
C:00013 cff1 .dw 0xcff1
C:00014 cff0 .dw 0xcff0
C:00015 e0f0 .dw 0xe0f0
C:00016 e4e0 .dw 0xe4e0
C:00017 e0d0 .dw 0xe0d0
C:00018 e6c0 .dw 0xe6c0
C:00019 e00a .dw 0xe00a
C:0001a 95c8 .dw 0x95c8
C:0001b 9209 .dw 0x9209
C:0001c 9631 .dw 0x9631
C:0001d 950a .dw 0x950a
C:0001e f7d9 .dw 0xf7d9
C:0001f cfe5 .dw 0xcfe5
C:00020 0100 .dw 0x0100
C:00021 0302 .dw 0x0302
C:00022 0504 .dw 0x0504
C:00023 0706 .dw 0x0706
C:00024 0908 .dw 0x0908
The first two lines in the output are the program memory ranges, which were excluded from disassembly. Why were these excluded? Because avrdis is a simple disassembler that can only follow the relative and absolute addresses from the branching instructions and does not try to perform semantic analysis of the code, or simulation of runtime behavior to infer possible code regions for disassembly.
The reason for this complexity comes from the fact that AVRs use a Modified Harvard Architecture which allows parts of the program memory to be accessed as data. This is very useful to store read-only data like character strings or data tables directly in the program memory. (Note that the SRAM is either absent, or rather limited in AVRs as opposed to the program memory.)
So since code and data can co-exist in the program memory and the interpreptation of data as code can lead to issues during disassembly when label addresses gets collected, avrdis uses a simple approach to disassemble the parts only, that are directly accessible from the branching instructions, thus guarantied to be code.
The parts which are potentionally data, gets emitted as .dw 0xnnnn. To enable the disassembly of such parts in case you're sure that those are code and not data, you can use the -e nnnn:nnnn option to specify a range. Multiple -e options are allowed to specify multiple ranges.
The full disassembly of a "mixed" firmware usually takes multiple iterations using the -l and -e options together, to explore the non-trivial parts.
After exploration of the above example, use the -l and -e options to get a listing of the properly disassembled code, with code words disassembled and data left as it is. The word address range 0x0020:0x0024 condains the data, that is referred by the ldi instructions at addresses C:00015 and C:00016 as decimal byte address high and low respectively. (Note that the word address 0x0020 translates to the byte address 0x0040 which is 0 high and 64 low in decimal.)
$ avrdis -l -e 4:4 foo.hex
0x000d:0x0024
C:00000 c004 rjmp L0
C:00002 9518 reti
C:00004 9518 reti
C:00005 b103 L0: in r16, 0x03
C:00006 7003 andi r16, 3
C:00007 2711 clr r17
C:00008 e0ed ldi r30, 13
C:00009 e0f0 ldi r31, 0
C:0000a 0fe0 add r30, r16
C:0000b 1ff1 adc r31, r17
C:0000c 9409 ijmp
C:0000d c003 .dw 0xc003
C:0000e c003 .dw 0xc003
C:0000f c003 .dw 0xc003
C:00010 c003 .dw 0xc003
C:00011 c003 .dw 0xc003
C:00012 cff2 .dw 0xcff2
C:00013 cff1 .dw 0xcff1
C:00014 cff0 .dw 0xcff0
C:00015 e0f0 .dw 0xe0f0
C:00016 e4e0 .dw 0xe4e0
C:00017 e0d0 .dw 0xe0d0
C:00018 e6c0 .dw 0xe6c0
C:00019 e00a .dw 0xe00a
C:0001a 95c8 .dw 0x95c8
C:0001b 9209 .dw 0x9209
C:0001c 9631 .dw 0x9631
C:0001d 950a .dw 0x950a
C:0001e f7d9 .dw 0xf7d9
C:0001f cfe5 .dw 0xcfe5
C:00020 0100 .dw 0x0100
C:00021 0302 .dw 0x0302
C:00022 0504 .dw 0x0504
C:00023 0706 .dw 0x0706
C:00024 0908 .dw 0x0908
Then reveal the ijmp targets.
$ avrdis -l -e 4:4 -e d:11 foo.hex
0x0020:0x0024
C:00000 c004 rjmp L0
C:00002 9518 reti
C:00004 9518 reti
C:00005 b103 L0: in r16, 0x03
C:00006 7003 andi r16, 3
C:00007 2711 clr r17
C:00008 e0ed ldi r30, 13
C:00009 e0f0 ldi r31, 0
C:0000a 0fe0 add r30, r16
C:0000b 1ff1 adc r31, r17
C:0000c 9409 ijmp
C:0000d c003 rjmp L1
C:0000e c003 rjmp L2
C:0000f c003 rjmp L3
C:00010 c003 rjmp L4
C:00011 c003 L1: rjmp L5
C:00012 cff2 L2: rjmp L0
C:00013 cff1 L3: rjmp L0
C:00014 cff0 L4: rjmp L0
C:00015 e0f0 L5: ldi r31, 0
C:00016 e4e0 ldi r30, 64
C:00017 e0d0 ldi r29, 0
C:00018 e6c0 ldi r28, 96
C:00019 e00a ldi r16, 10
C:0001a 95c8 L6: lpm
C:0001b 9209 st Y+, r0
C:0001c 9631 adiw r31:r30, 1
C:0001d 950a dec r16
C:0001e f7d9 brne L6
C:0001f cfe5 rjmp L0
C:00020 0100 .dw 0x0100
C:00021 0302 .dw 0x0302
C:00022 0504 .dw 0x0504
C:00023 0706 .dw 0x0706
C:00024 0908 .dw 0x0908
Finally, the raw source code can be redirected to a .asm file for further tinkering in a code editor.
$ avrdis -e 4:4 -e d:11 foo.hex >foo.asm
$ cat foo.asm
.org 0x0000
rjmp L0
.org 0x0002
reti
.org 0x0004
reti
L0: in r16, 0x03
andi r16, 3
clr r17
ldi r30, 13
ldi r31, 0
add r30, r16
adc r31, r17
ijmp
rjmp L1
rjmp L2
rjmp L3
rjmp L4
L1: rjmp L5
L2: rjmp L0
L3: rjmp L0
L4: rjmp L0
L5: ldi r31, 0
ldi r30, 64
ldi r29, 0
ldi r28, 96
ldi r16, 10
L6: lpm
st Y+, r0
adiw r31:r30, 1
dec r16
brne L6
rjmp L0
.dw 0x0100
.dw 0x0302
.dw 0x0504
.dw 0x0706
.dw 0x0908