Quantum,
Thanks for reposting, I am sure it will help many!
I had been preparing for this operation for a long time, but then had to perform it urgently in the end of Nov, and then had to redo much of the research from scratch. (Including finding the ethernet pinout error in the Tesla schematics that I see corrected in the image above
)
Some of my experiences for the benefit of others:
The amount of bit-errors in the already failing eMMC increased massively after de-soldering (probably due to the heat exposure), but otherwise it read fine in the AllSocket.
Fortunately I had managed to read some images using fly-leads and an ISP, so doing that is advisable!
Sadly it was very hard to get an in-circuit readout to work using either the AllSocket adapter, or a 1-bit capable SD card reader (Transcend RDF5) with custom adapters. In the end we had to cut D0 to the CPU (transmission line reflections are a bitch), short out all crystals on the Tegra board (to halt the CPUs), and run the board at full 3.3V voltage to make it work...
I used a 16GB Swissbit EM-26, preconfigured for extended reliability (Effectively halving the usable size). That worked fine in the CID. (But be careful if doing this to the CI, as that allegedly has an older kernel that might not support eMMC 5.0). If using a low level programmer, burn also the (writable bits in the) EXT_CSD area, as copied from the original chip.
I wrote some custom tools to merge together a working image from the bits and pieces recovered, but otherwise unsquashfs, diff -r, and fsck are your friends, and should hopefully be enough...
Thanks for reposting, I am sure it will help many!
I had been preparing for this operation for a long time, but then had to perform it urgently in the end of Nov, and then had to redo much of the research from scratch. (Including finding the ethernet pinout error in the Tesla schematics that I see corrected in the image above
)Some of my experiences for the benefit of others:
The amount of bit-errors in the already failing eMMC increased massively after de-soldering (probably due to the heat exposure), but otherwise it read fine in the AllSocket.
Fortunately I had managed to read some images using fly-leads and an ISP, so doing that is advisable!
Sadly it was very hard to get an in-circuit readout to work using either the AllSocket adapter, or a 1-bit capable SD card reader (Transcend RDF5) with custom adapters. In the end we had to cut D0 to the CPU (transmission line reflections are a bitch), short out all crystals on the Tegra board (to halt the CPUs), and run the board at full 3.3V voltage to make it work...
I used a 16GB Swissbit EM-26, preconfigured for extended reliability (Effectively halving the usable size). That worked fine in the CID. (But be careful if doing this to the CI, as that allegedly has an older kernel that might not support eMMC 5.0). If using a low level programmer, burn also the (writable bits in the) EXT_CSD area, as copied from the original chip.
I wrote some custom tools to merge together a working image from the bits and pieces recovered, but otherwise unsquashfs, diff -r, and fsck are your friends, and should hopefully be enough...
