Jump to content
Click here to contact our helpful office staff ×

Recommended Posts

Posted

Download Link: http://andrewrevill.co.uk/Downloads/MEMS3Tools.zip

NOTE: THIS DOCUMENTATION IS UP TO DATE AS OF VERSION 4.90 RELEASE OF THE MEMS3 TOOLS APP

In the latest releases of my MEMS3 Mapper tool I have gone to great lengths to try to ensure that ECUs cannot be accidentally bricked by writing bad firmware or map to the ECU. However, there are some things that I just cannot protect against, for example the user manually editing the memory bytes in the hexadecimal editor to produce something that is not valid, or writing a firmware to hardware that just cannot support it. The small but non-zero risk of bricking an ECU has always bothered me. Whilst researching these ECUs I’ve bricked many, and the only way to recover them so far has been to open them up, de-solder the main EEPROM chip using hot air, reprogram the EEPROM with valid firmware and map using an EEPROM programmer on the bench and then solder the chip back in, which is tedious and time-consuming problem and not exactly good for the board and other components.

I always suspected that the engineers who designed it would have left a back door in somewhere to allow a bricked ECU to be recovered.

Bricked ECUs

In addition to the firmware and map, the EEPROM chip contains a boot loader program. When a brand new virgin ECU is suppled, the boot loader is all that is present. The boot loader has sufficient functionality to allow the ECU to communicate with a programmer and load the firmware and map. The boot loader generally checks to see whether valid firmware and map are present and if so, shortly after the ECU boots it transfers control straight to the firmware. The boot loader initialises watchdog timers which reboot the ECU if the firmware appears to have hung up, and what generally seems to happen in the case of a bricked ECU is that the boot loader sets up the watchdogs, transfers control to the firmware which fails to operate normally, the watchdogs detect this and reboots the ECU which then just repeats the same cycle. The ECU is stuck in an indefinite boot loop. You usually see the fuel pump starting to prime then being switched off again several times a second forever. Because the ECU is constantly rebooting, you never get the chance to establish stable communications with it so cannot even begin to reprogram it with valid firmware and map.

The boot loader code is protected and permanent. It is loaded into the first sector of the EEPROM chip at manufacture and the ECU does not appear to provide any method to erase or modify it. Certainly my MEMS3 Mapper and Flasher tools never touch the boot loader in an ECU. This means that whatever we do to the firmware and map, however badly we screw those up to brick the ECU, the boot loader will still be clean and valid and just the way it was the day the ECU was manufactured.

If only I could find some way to prevent the boot loader from transferring control to the firmware, as though no firmware had been loaded …

Back Door Search

I decided to search for a possible back door by working through a disassembly of the boot loader code. Now I’ve spent quite a long time looking at disassembled code from the ECU in the past and I have to say it’s not been anywhere near as successful as I would have hoped. The code is very opaque and difficult to get into. But I thought that if I was searching for the answer to a very specific question I might stand a chance, and in this case I had few pointers to help me get started. For example, once the boot loader jumps into the firmware, the firmware mostly executes as a separate stand-alone program. The boot loader should be independent of the firmware, so there are very few references to firmware addresses in the boot loader code and any references that I did find were likely to be related to the process of checking for loaded firmware and launching the firmware.

Back Door Analysis

The analysis below is based on a disassembly of the VVC 160 ECU NNN000160’s boot loader “bootp033” but the code varies very little across different boot loaders.

Subroutine $001007CA is responsible for jumping into the firmware (at addresses $001007D8$001007DE and $001007E2). It loads the address of the firmware $110000 into register a0, offsets that by 4 bytes and reads the vector at that address to put the firmware entry point into register a0, then jumps to the address that register a0 points to:

ROM:001007CA
ROM:001007CA ; =============== S U B R O U T I N E =======================================
ROM:001007CA
ROM:001007CA
ROM:001007CA sub_1007CA:                             ; CODE XREF: ROM:00100622↑p
ROM:001007CA                 btst    #0,d1
ROM:001007CE                 bne.s   loc_1007E6
ROM:001007D0                 cmpi.l  #0,d0
ROM:001007D6                 bne.s   loc_1007E6
ROM:001007D8                 movea.l #dword_110000,a0
ROM:001007DE                 movea.l 4(a0),a0
ROM:001007E2                 jmp     (a0)
ROM:001007E4 ; ---------------------------------------------------------------------------
ROM:001007E4                 bra.s   locret_1007EA
ROM:001007E6 ; ---------------------------------------------------------------------------
ROM:001007E6
ROM:001007E6 loc_1007E6:                             ; CODE XREF: sub_1007CA+4↑j
ROM:001007E6                                         ; sub_1007CA+C↑j
ROM:001007E6                 bsr.w   sub_1007F8
ROM:001007EA ; ---------------------------------------------------------------------------
ROM:001007EA
ROM:001007EA locret_1007EA:                          ; CODE XREF: sub_1007CA+1A↑j
ROM:001007EA                 rts
ROM:001007EA ; End of function sub_1007CA
ROM:001007EA

Before executing the firmware, it performs a couple of tests. Specifically it tests that both bit 0 of register d1 is 0 (at addresses $001007CA and $001007CE) and register d0 contains 0 ($001007D0 and $001007D6) when the subroutine is called. The firmware is only executed if both of these tests are passed.

Subroutine $001007CA above is called right at the end of the main entry point routine of the boot loader. Before calling $001007CA this calls other boot loader subroutines $00100628$00100708$00100740 and $001007A2:

ROM:00100612
ROM:00100612 loc_100612:                             ; CODE XREF: ROM:00100576↑j
ROM:00100612                 bsr.w   sub_100628
ROM:00100616                 bsr.w   sub_100708
ROM:0010061A                 bsr.w   sub_100740
ROM:0010061E                 bsr.w   sub_1007A2
ROM:00100622                 bsr.w   sub_1007CA
ROM:00100626
ROM:00100626 locret_100626:                          ; CODE XREF: ROM:00100610↑j
ROM:00100626                 rts
ROM:00100628

Subroutine $00100740 checks for the $5AA5 signatures of the firmware (at firmware address $110410, in code starting at address $00100752) and map (at map address $13C012, in code starting at address $0010075E) (you can look at these addresses in any valid firmware and map using MEMS3 Mapper and see that they do indeed always contain the signature bytes $5AA5) and sets bits in register d0 if they are not found, so the check for register d0 being 0 amounts to a check that firmware and map are loaded.

 

ROM:00100740
ROM:00100740 ; =============== S U B R O U T I N E =======================================
ROM:00100740
ROM:00100740
ROM:00100740 sub_100740:                             ; CODE XREF: ROM:0010061A↑p
ROM:00100740                 movea.l #dword_110000,a0
ROM:00100746                 clr.l   d0
ROM:00100748                 cmpi.w  #0,d7
ROM:0010074C                 bne.s   loc_100752
ROM:0010074E                 bset    #1,d0
ROM:00100752
ROM:00100752 loc_100752:                             ; CODE XREF: sub_100740+C↑j
ROM:00100752                 cmpi.w  #$5AA5,$410(a0)
ROM:00100758                 beq.s   loc_10075E
ROM:0010075A                 bset    #0,d0
ROM:0010075E
ROM:0010075E loc_10075E:                             ; CODE XREF: sub_100740+18↑j
ROM:0010075E                 movea.l #word_13C000,a0
ROM:00100764                 cmpi.w  #$5AA5,$12(a0)
ROM:0010076A                 beq.s   loc_100770
ROM:0010076C                 bset    #2,d0
ROM:00100770
ROM:00100770 loc_100770:                             ; CODE XREF: sub_100740+2A↑j
ROM:00100770                 movea.l #unk_110400,a1
ROM:00100776                 movea.l #word_13C00A,a0
ROM:0010077C                 move.b  #0,d2
ROM:00100780                 bra.s   loc_100786
ROM:00100782 ; ---------------------------------------------------------------------------
ROM:00100782
ROM:00100782 loc_100782:                             ; CODE XREF: sub_100740+5E↓j
ROM:00100782                 addi.b  #1,d2
ROM:00100786
ROM:00100786 loc_100786:                             ; CODE XREF: sub_100740+40↑j
ROM:00100786                 cmpi.b  #7,d2
ROM:0010078A                 bgt.s   locret_1007A0
ROM:0010078C                 move.b  (a1),d1
ROM:0010078E                 cmp.b   (a0),d1
ROM:00100790                 beq.s   loc_100796
ROM:00100792                 bset    #3,d0
ROM:00100796
ROM:00100796 loc_100796:                             ; CODE XREF: sub_100740+50↑j
ROM:00100796                 adda.w  #1,a0
ROM:0010079A                 adda.w  #2,a1
ROM:0010079E                 bra.s   loc_100782
ROM:001007A0 ; ---------------------------------------------------------------------------
ROM:001007A0
ROM:001007A0 locret_1007A0:                          ; CODE XREF: sub_100740+4A↑j
ROM:001007A0                 rts
ROM:001007A0 ; End of function sub_100740
ROM:001007A0
ROM:001007A2

So we now know that the firmware will be executed if both firmware and map are loaded and bit 0 of register d1 is 0 when subroutine $001007CA is called. We also know from previous experience that when an ECU does not have firmware or map loaded, it happily runs the boot loader and allows normal programming over OBDII using the MEMS3Mapper application. Looking at the code, I would expect identical behaviour if the other test failed, i.e. if bit 0 of register d1 is 1 when subroutine $001007CA is called. So if we can find a way of setting this bit at this point, we should have a way of preventing the ECU from executing the firmware and accepting programming just as though it was a virgin ECU, and that would be the back door we were looking for.

So what is bit of 0 of register d1 doing?

Subroutine $001007A2, which is called in the boot loader immediately before the signature checking routine $00100740 above, manipulates bit 0 of d1, and so it determining the other requirement for the firmware to be executed. It’s a very simple subroutine which sets bit 0 of d1 entirely based on registers of the QSM (Queued Serial Module) in the MC68336 microcontroller:

ROM:001007A2
ROM:001007A2 ; =============== S U B R O U T I N E =======================================
ROM:001007A2
ROM:001007A2
ROM:001007A2 sub_1007A2:                             ; CODE XREF: ROM:0010061E↑p
ROM:001007A2                 btst    #6,($FFFFFC0D).w
ROM:001007A8                 beq.s   loc_1007C4
ROM:001007AA                 cmpi.b  #$96,($FFFFFC0F).w
ROM:001007B0                 bne.s   loc_1007C4
ROM:001007B2                 bset    #0,d1
ROM:001007B6                 bset    #0,($FFFFFC0B).w
ROM:001007BC                 bclr    #0,($FFFFFC0B).w
ROM:001007C2                 bra.s   locret_1007C8
ROM:001007C4 ; ---------------------------------------------------------------------------
ROM:001007C4
ROM:001007C4 loc_1007C4:                             ; CODE XREF: sub_1007A2+6↑j
ROM:001007C4                                         ; sub_1007A2+E↑j
ROM:001007C4                 bclr    #0,d1
ROM:001007C8
ROM:001007C8 locret_1007C8:                          ; CODE XREF: sub_1007A2+20↑j
ROM:001007C8                 rts
ROM:001007C8 ; End of function sub_1007A2
ROM:001007C8
ROM:001007CA

If bit 6 of QSM register SCSR is 0 (tested at addresses $001007A2 and $001007A8) then bit 0 of register d1 is cleared to 0 (at address $001007C4), otherwise if the low byte of QSM register SCDR is not $96 (tested at addresses $001007AA and $001007B0) then bit 0 of register d1 is also cleared to 0 (at address $001007C4), otherwise bit 0 of d1 is set to 1 (at address $001007B2), bit 0 of QSM register SCCR1 is toggled to 1 then 0. The subroutine then exits in all cases.

Now the SCSR register is the SCI Status Register and SCDR register is the SCI Data Register, and the SCI is the Serial Communications Interface which is part of the QSM. It is basically a standard UART serial port. SCSR tells us about the current status of port operation and SCDR holds the most recent byte received. From the MC68336 data sheet, Bit 6 of the status register SCSR is the RDRF flag, or Receive Data Register Full – it tells us if there’s something in the data register SCDR.

So the condition which determines that a loaded firmware should be executed is simply that no byte has been received by the SCI port, or that byte was not $96.

In other words, we can prevent any loaded firmware being executed by sending the byte $96 to the SCI port at the right moment.

That leaves three questions:

1)What is the right moment?

2)What is the SCI port connected to?

3)What Baud rate should we transmit the $96 at?

The answer to question 1 is clearly “as the ECU boots”, but exactly when during the boot sequence is not easy to determine, so I decided to try the approach of just broadcasting the byte $96 continuously as the ECU was powered on. I couldn’t be sure this would work but it was certainly worth a try.

The answer to question 2 seemed likely to be the OBDII serial port. My guess was that SCI was just connected to the OBDII K-Line as this was used for serial UART communications with the ECU. It would be extremely convenient if this turned out to be the case as I could provide a utility within MEMS3 Mapper that just broadcast $96 to the OBDII port while you powered the ECU on.  Again, I couldn’t be sure this was the case without a lot more digging in the firmware disassembly or tracing the electronics, but the easiest thing was to give it a try.

As for question 3, one of the subroutines called immediately prior to those discussed above by the boot loader was $00100628, and that contained the following lines:

ROM:0010069E
ROM:001006A0 ; ---------------------------------------------------------------------------
ROM:001006A0
ROM:001006A0 loc_1006A0:                             ; CODE XREF: ROM:0010051A↑j
ROM:001006A0                 move.w  #$34,($FFFFFC08).w ; '4'
ROM:001006A6                 move.w  #$C,($FFFFFC0A).w
ROM:001006AC                 bra.l   loc_100520
ROM:001006B2 ; ---------------------------------------------------------------------------
ROM:001006B2

These load the SCCR0 register of the QSM with the value $34 (at address $001006A0). SCCR0 is the SCI Control Register 0 and controls the Baud rate at which the SCI port operates, according to the formula Baud Rate = (Clock Frequency) / 32 / SCCR0. So with a system clock frequency of 16Mhz and an SCCR0 value of $34 or 52 decimal, the Baud rate was configured to 16000000 / 32 / 52 = 9615, which is as close as you can get to the standard OBDII Baud rate of 9600 Baud. This provided even more evidence that I was on the right lines thinking that the SCI port was actually the OBDII K-Line.

So my plan was to broadcast $96 at 9600 Baud over the OBDII port while booting up the ECU and … rather to my surprise, it worked! That’s the back door I was looking for.

So that provides a failsafe way of being able to recover a bricked ECU. I’ve tested using it to prevent a perfectly healthy ECU from executing the firmware - the ECU just sits in boot / programming / recovery mode in the boot loader waiting to be programmed. I’ve tested communications with the ECU after blocking the firmware with this signal and all seems normal. Finally I’ve used to unbrick a development ECU that got bricked while I was playing with it. I’m pretty much 100% sure that this method will unbrick any MEMS3 ECU that has bad firmware or map as the boot loader code is protected and always virgin. It doesn’t matter how the ECU was bricked, whether it was using MEM3 Flasher / Mapper or other programming tools such as Galletto, this method provides a safe recovery option.

Recover Bricked ECU

In release 4.87 of MEMS3 Flasher and MEMS3 Mapper I’ve now added an option to the ECU Tools menu to Recover Bricked ECU. When you select this option, the application will begin to broadcast the request code at regular intervals and monitor for the expected response. If the ECU is stuck in a boot loop, constantly rebooting, it should recover almost immediately as the next time it reboots it will detect the request and remain in the boot loader ready to accept programming. If the response is not detected within a short time, the following dialog is displayed:

image001.png

If the ECU is hanging and unable to communicate, we need to persuade it to reboot. Switching the ignition off for at last 15 seconds (and then back on again) is usually sufficient to ensure that it will perform a full boot when powering on again. At this point it should again remain in the boot loader ready to accept programming as above. If the ECU was really tightly stuck in a loop, then just occasionally switching the ignition off is not sufficient to break it out of the loop and trigger a reboot. In this case, you need to briefly remove power from the ECU. When power is re-applied and the ignition is turned on it will boot again, detect the request and remain in the boot loader ready to accept programming as above.

Depending on the timing of the request code, it may occasionally require more than one attempt to recover an ECU.

When the expected response is detected, the following dialog is displayed:

image003.png

Once the ECU has recovered into a state where it is running the boot loader and willing to communicate with the application again it is important that you then write a good replacement firmware, map or both (depending on what was damaged, if in doubt do both). The initial recovery is only temporary; we have broken the ECU out of the cycle that was preventing it from communicating and accepting programming using the special code, but next time it reboots without seeing the request broadcast it will of course continue to try to execute the damaged firmware or map as before. In recovery mode, some of the ECU tools and operations will fail (the ECU will reject the operations as it will not be running the firmware) and it will not provide any engine management functionality, but everything necessary to read and write the firmware and map will function normally.

Once you have rewritten the ECU the repair will be permanent.

  • Area Representative
Posted
School! It wouldn't have mattered how much attention I paid, ECUs & injection wouldn't have featured. They were only just moving from external combustion engines when I was at school (don't misunderstand, triple-expansion had been invented!)
Posted

I've done a little bit more digging with my MEMS3 Terminal applications, watching the communications between my MEMS3 Mapper and the ECU. It seems the ECU responds to the $96 code with a $00 byte when it abandons loading the firmware and remains in the boot loader.

Armed with this knowledge I've been able to tidy up the user interface a bit. Now if the ECU is stuck in a boot loop (the most common failure mode) it will just recover immediately without prompting the user to do anything. Only if it doesn't receive the $00 response within a short time (200ms) does it now prompt the user to turn the ignition off, or failing that to power down the ECU. The dialog closes automatically when it detects that the ECU has recovered (it now only has a Cancel button and no OK) and the About ECU dialog displayed contains a large banner confirming that it's in recovery mode and you need to write firmware and map to complete the repair.

It took a little more thinking about as one of the suggested recovery actions is to power down the ECU; this can lead to the OBDII K-Line being pulled to a low state and this looks like a $00 byte being sent by the ECU. What I've done is to have two threads running, one writing the $96 codes and one monitoring for the $00 response asynchronously. The monitoring thread sees everything on the line, anything sent by the ECU and anything sent by the application. In the case of a genuine $00 response from the ECU it will therefore see $00 followed very shortly by the next $96 sent by the writing thread. In the case of the ECU being briefly powered down, if this does look like a spurious $00 it will see $00 then ... silence. So it only accepts a $00 followed within 100ms my a $96 as a genuine ECU recovery response. In reality the ECU seems to response within 1ms, and the 10ms gaps between transmissions of the $96 code ensure no conflicts. This seems to work really well.

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...