Language: English · Русский
Some evocode changes can be rehearsed in an emulator before you touch a real head unit. IDEmu is a QEMU-based emulator of the NBT EVO / iDrive6 head unit (macOS) that boots the unit's own software from your own PSdZData/SWFL, with SSH and a coding editor. (License: PolyForm Noncommercial.)
The one thing it is genuinely useful for here is the blobB anchor swap — confirming that a patched
NBTCarHU (evocode fsc patch-blobb) boots the unit's own software without bricking it, which is
the scariest step of the on-car FSC procedure.
| ✅ Rehearsable in IDEmu | ❌ Only on the car |
|---|---|
A patched NBTCarHU boots QNX (services + SSH come up) |
Diagnostic-plane commands — write, fsc apply/read, root — IDEmu does not expose the HSFZ endpoint (TCP 6801) |
| No boot regression vs. stock (reaches the same stage) | Whether an FSC is actually accepted (needs the diagnostic write + a running NBTCarHU) |
| The binary stays intact / length-preserving | HMI render and feature activation — validated only for specific HMI builds¹ |
| Coding survives (it lives in a separate image, untouched by a blobB swap) | Maps / navigation, Android Auto |
¹ If your HMI build is not in IDEmu's validated set, boot stalls near the end (the HMI never finishes)
and NBTCarHU — gated behind HMI readiness — will not start. That is an emulator/version limit, not
your patch: you still validate the system layer (boot / SSH / coding). Compare patched vs. stock —
if both reach the same stage, the swap is boot-safe.
- Baseline (stock). Build a version-matched environment from your own SWFL, cold-boot it, and over
SSH record:
uname -a, the size ofNBTCarHU,pidin a | grep -i carhu(which processes are up), and how far boot got. This separates "my patch broke it" from "the emulator behaves this way on my build." - Build the patched binary with evocode (length-preserving — only the 128-byte blobB changes):
evocode fsc find-blobb NBTCarHU.stock # offset + current anchor fingerprint (per-build — do not hardcode) evocode fsc patch-blobb NBTCarHU.stock --key owner.json --backup stock_blobB.bin -o NBTCarHU.patched - Inject into the image. Because the patch is length-preserving, the stock 128-byte anchor appears
verbatim in the environment's root image — replace it in place, then cold-boot (the root image is
a RAM-disk, so edits inside a running guest do not survive a reboot):
(Alternative, closer to the on-car step:
img = bytearray(open(".../sda.img", "rb").read()) old = open("stock_blobB.bin", "rb").read() # from --backup above new = open("owner_blobB.bin", "rb").read() # your anchor (evocode fsc keygen --blobb) i = img.find(old) assert i >= 0 and img.find(old, i + 1) == -1, "anchor not found / not unique in image" img[i:i + 128] = new open(".../sda.img", "wb").write(img)
scpthe patchedNBTCarHUinto a running guest and restart the HMI — it mirrors the at-car rename-over, but it is not a disk cold-boot.) - Cold-boot the patched image and verify against the baseline.
| Check | PASS | FAIL |
|---|---|---|
| Cold-boot of the patched image | reaches the baseline stage, QNX + services alive | panics / hangs earlier than baseline |
| SSH into the guest | answers; uname = QNX OMAP5 |
no session |
NBTCarHU process |
present in pidin a, not crash-looping |
absent / crash-loop |
| Binary size | unchanged (patch is length-preserving) | changed |
| Anchor in the image | your anchor's fingerprint (evocode fsc fingerprint) |
stock / foreign fingerprint |
PASS = the anchor swap is boot-safe → on the car you risk only the write transport, not a brick.
It rehearses the firmware-file install (that the patched NBTCarHU boots). It does not cover
the diagnostic-plane steps (fsc apply/read, coding write, root enable — no HSFZ in the
emulator), feature verification, maps, or Android Auto. Those stay for the car. See
root-access.md for the firmware/filesystem plane and ../AGENTS.md §1
for the two-plane model.
- Find the blobB offset with
evocode fsc find-blobb— never hardcode it; it differs per build. - Single-core TCG emulation is slow; let boot run in the background, do not poll.
- Standing up a stand (build IDEmu, your own version-matched SWFL + the version-pin gotcha, create/build/run, the console bridge's quirks, and what each kind of stand can develop) is in dev-stand.md.
- An IDEmu console bridge forwards a shell but does not pipe exec stdin (confirmed:
cat-over-exec writes 0 bytes), so it can't exerciseevocode fw's file transfer — only its choreography (remount / backup / rename-over / chmod on real QNX, viafw --modern).fw installverifies the on-unit byte-count and aborts on mismatch, so that limitation — or a dropped transfer — can't install an empty file.