Two CAN controllers on a 64-pin microcontroller leave very little room for anything else. The LPC2119FBD64 is the 128 KB member of the LPC2100 line with those two channels, an acceptance filter shared between them, 16 KB of SRAM and a four-channel 10-bit ADC. It was built for gateways, and twenty years later most of the boards still in service have no source tree left. Recovering a protected LPC2119FBD64 binary therefore starts with a question that has nothing to do with the debugger: which parts of the application are in flash at all?
Acceptance Filters Are Not Firmware
The CAN controllers keep their message objects and filter tables in RAM and in dedicated registers. None of that appears in a flash dump. A customer who expects the recovered image to explain why a gateway forwards one identifier and drops another will be disappointed unless someone also documents the filter configuration. We separate the two from the outset: the code and constants live in the array, while the runtime state lives elsewhere and has to be reconstructed from the application itself.
That distinction is easy to miss on a part with this much peripheral integration. Two UARTs, two SPI ports, a fast I2C bus, six PWM outputs, two 32-bit timers and a vectored interrupt controller all feed the same 16 KB of SRAM. A readout gives you the logic that configures them, not the configuration captured at a particular moment.
Sector Zero Carries Two Different Things
To recover protected LPC2119FBD64 binary, start with the sector map.
The flash array is organised in 4 KB sectors at the bottom and 32 KB sectors above, with 4 KB sectors again near the top. Sector 0 holds the ARM exception vectors, including the reserved word at 0x14 whose value must make the eight vectors sum to zero, and it also holds the Code Read Protection word at 0x1FC. One 4 KB erase therefore removes both the entry point of the application and the setting that governs whether it can be read.
That is why the CRP level changes the geometry of a recovery. CRP1 forbids writing sector 0 but still allows individual sector erases. CRP2 disables read, write-to-RAM, copy-RAM-to-flash and go commands, leaving only a full-chip erase. CRP3, when valid user code exists, closes the P0.14 ISP entry pin as well. Any route that begins with an erase has already given up the objective.
Bootloader Revision Comes First
Early bootloader code implemented a single protection level selected by 0x87654321. Later /00 and /01 silicon added the three-level scheme with 0x12345678, 0x87654321 and 0x43218765, and NXP documented the diversified levels as available on the /00 parts too. Two boards with identical markings can behave differently at the serial prompt. Reading the boot code version over ISP is cheap and decides the whole plan, so it is the first electrical step after the marking is confirmed.
CRP changes also require a complete power cycle. A freshly programmed board can still respond as an unprotected device until the supply has been removed and restored. We log the revision and the observed behaviour before any erase is considered.
The Physical Route
When the boot ROM offers no read command, the array has to be approached at the die. An LQFP64 with a 10 by 10 millimetre body gives a modest access window, and the flash controller’s check bytes have to be located before the user data can be interpreted. Reads are taken in sector-sized blocks and repeated from a second angle, then compared. A single-bit error corrected by the on-chip logic may be invisible in a raw capture, but it will change an instruction word.
The vector table is rebuilt at 0x00000000 with the checksum word recomputed, because the boot ROM will not start an image whose vectors do not sum to zero. Thumb-heavy builds need particular care during disassembly: switching between 16-bit and 32-bit instruction streams at subroutine boundaries is normal on ARM7, and a flat 32-bit decode produces convincing nonsense.
What A Gateway Board Asks For
Most requests concern a unit that still works but cannot be reproduced. The useful answer is not just a binary; it is a binary, a sector map, the CRP state actually found, and a note on which peripheral registers must be re-derived from the code. For a part that shares its flash layout with the LPC2114, see that variant’s readout notes. Our contact route is where board photographs and part markings go, the memory-type comparison explains why this array is neither OTP nor mask ROM, and this ARM microcontroller case covers a related 32-bit recovery. The LPC2132FBD64 page discusses a flash array with error correction.