bootloader was not packed correctly (v4.0.0-rc3) for Sengled G4

[hlyi]

@jnilo1 I tried to use BOARD=sengled-e39-g8c flash_install_rtl8196e.sh to update sengled G4 to v4.0.0-rc3. i=It seems that the script packed wrong bootloader, and caused soft brick of the unit. Could you please check if sengled-e39-g8c bootloader was packed properly? Thanks

[jnilo1]

What happened — confirmed, and sorry about the brick

You are right, and I want to be transparent about the mechanism. The repo ships one pre-built bootloader binary, 3-Main-SoC-Realtek-RTL8196E/31-Bootloader/boot.bin, and it is the Lidl build (V2.9). flash_install_rtl8196e.sh assembles the full-flash image from that file regardless of BOARD= — for a non-lidl board it only prints a warning on stderr:

Warning: BOARD=sengled-e39-g8c — make sure 31-Bootloader/boot.bin was built with
  'BOARD=sengled-e39-g8c ./build_bootloader.sh'; a mismatched bootloader bricks the gateway (DRAM).

So the image you flashed contained the Lidl bootloader, whose DRAM bring-up is wrong for the G4 — hence the soft brick. The board-mismatch guard (device-tree model check) could not catch this: it validates that the BOARD= you pass matches the board you are flashing, but the bootloader binary itself carries no board identity we can inspect.

A warning was clearly not enough of a guard here. My apologies for the lost time.

Correct procedure today (v4.0.0-rc3)

Build the G4 bootloader first, then run the installer:

cd 3-Main-SoC-Realtek-RTL8196E/31-Bootloader
BOARD=sengled-e39-g8c ./build_bootloader.sh        # overwrites boot.bin with the G4 build

cd ../..
BOARD=sengled-e39-g8c ./flash_install_rtl8196e.sh

For anyone else landing here after the same mistake: recovery is via the serial console and the first-install procedure in 3-Main-SoC-Realtek-RTL8196E/35-Migration/. @hlyi — I assume you managed to recover since you have been testing OT-RCP since; please confirm, and tell me if the recovery path gave you any trouble.

Fix planned for v4.0.0

The bootloader build is reproducible, so the Lidl boot.bin has a stable checksum. The full-flash scripts will hard-fail (instead of warning) when BOARD≠lidl and boot.bin is still the Lidl binary. Longer term, per-board pre-built bootloaders with a board tag embedded in the binary are on the roadmap, but the hard guard closes the brick scenario now.

One heads-up on timing: we will not publish any new release candidate until issue #99 is definitively closed (rc4 is under field validation), so this fix will land with v4.0.0 itself.

[hlyi]

@jnilo1 No worries about the soft brick. Thanks for supporting Sengled G4 porting. Yes, I was able to recover using bootloader mode to reflash full image ( with correct Sengled G4 bootloader binary). Some background information: before v4.0.0-rc3, I used locally built binary for Sengled G4. When trying v4.0.0-rc3, I checked out new workarea, and used your pre-built image. I think what happened was that the bootloader only initializes 32M memory (DDR timing may be incorrect, but at least good enough for bootloader to work), and kernel (using dts) assumed 64M memory and caused malfunction.

[jnilo1]

Fix implemented — and it went further than the md5 guard

Thanks for confirming the recovery. Your analysis matches what the code says: the Lidl stage-1 DDR values are close enough for the bootloader's modest needs on the G4 but wrong for its 64 MiB DDR2, so things fall over once the kernel (whose DTS declares the full 64 MiB) starts using memory the controller never brought up correctly.

Rather than the md5 hard-fail I promised above, I implemented the "longer term" item directly — the bootloader now uses the exact same per-board pre-built mechanism as the kernel images:

• One committed pre-built per board: 31-Bootloader/boot-img/<board>/boot.bin (lidl and sengled-e39-g8c).
• Every consumer — build_fullflash.sh, create_fullflash.sh, flash_install_rtl8196e.sh, flash_remote.sh bootloader, flash_bootloader.sh — resolves the binary from BOARD= (default lidl), through the same shared helper that resolves the kernel image. flash_install_rtl8196e.sh fails fast, before touching the gateway, if the pre-built for your BOARD= is missing.
• build_bootloader.sh now writes only into the selected board's slot, so there is no shared mutable boot.bin any more. That closes the scenario you hit (Lidl bring-up bundled into a G4 fullflash) and also the mirror one (a leftover G4 build silently landing on a Lidl gateway at the next default flash).
• flash_remote.sh bootloader additionally gained the /proc/device-tree/model board-mismatch guard that previously covered only the kernel component.

So with this in place, BOARD=sengled-e39-g8c ./flash_install_rtl8196e.sh does the right thing out of the box — no manual bootloader build step first.

One zero-risk check you can do for us

The bootloader build is deterministic (pinned build timestamp), so two builds of the same source are byte-identical. The committed G4 pre-built, compiled from the same V2.9 source that is in your v4.0.0-rc3 checkout, has:

md5sum boot-img/sengled-e39-g8c/boot.bin
b43b342db3b2410d6d7a87c7d74985b5

Could you run md5sum on the G4 boot.bin you built in your rc3 workarea (the one your recovered unit is running)? If it matches, the committed pre-built is bit-identical to a binary that is field-validated on your hardware, and we can ship it in v4.0.0 with full confidence. If it differs, tell me which tree/commit and toolchain you built with so I can account for the delta (a pre-V2.9 checkout, or a compiler other than the one in the project's Docker image, would legitimately produce different bytes).

If you no longer have that build around, or you'd rather verify the committed binary directly, say so and I'll attach it here (zipped — GitHub doesn't take raw .bin) for you to flash from bootloader mode.

Timing

As said above: this lands publicly with v4.0.0 — no new release candidate before issue #99 is definitively closed (rc4 is still under field validation).

[hlyi]

@jnilo1 Thanks a lot for the quick fix. Yes, boot.bin I created on v4.0.0-rc4 tag has the same md5 hash as your binary, and it works fine on my Sengled G4 unit.

md5sum 3-Main-SoC-Realtek-RTL8196E/31-Bootloader/boot.bin 
b43b342db3b2410d6d7a87c7d74985b5  3-Main-SoC-Realtek-RTL8196E/31-Bootloader/boot.bin

[hlyi]

Thanks for fixing it in v4.0.0.-rc5. Close the discussion