Goldshell – Firmware Bricked (Red & Green LEDs Stuck)

Power-cycle and wait 5 full minutes. Unplug the 12V PSU. Unplug Ethernet. Wait 30 seconds. Re-plug PSU only. Do not plug Ethernet yet. Watch both LEDs for 5 minutes. Some "bricks" are mid-boot after a bad shutdown - the internal filesystem check (`fsck`) runs silently and only then advances the LEDs. If after 5:00 the LEDs haven't changed state, proceed. If green comes on solid or red+blue appears (pool-connect), you were never bricked.

Email `hello@goldshell.com` with your model + serial requesting the recovery image. Subject line: `burn-<model>.img firmware recovery request - SN: <serial>`. Include: exact model (e.g. "KD-BOX Pro", "Mini-DOGE II", "ST-BOX"), serial number from the rear sticker, the firmware version that was being flashed, and what happened. Response is typically 24-72 hours. This is the only documented source for the recovery image - Goldshell publishes normal firmware but not `burn-*.img` on their site or GitHub.

Factory reset via the RST pinhole (low probability, worth trying). Miner powered, fan running. Insert a paperclip or SIM-eject tool into the `RST` pinhole on the back. Hold for 5-10 seconds until the LEDs blink or change. Release. Wait 2 minutes. This rarely fixes a true firmware brick but costs nothing to try. If the LEDs still stay stuck after the 2-minute wait, the reset partition either isn't populated (common on older units) or the corruption is deep enough that the reset config can't load.

Order a quality microSD card if you don't have one handy. 8 GB or 16 GB, Samsung EVO Plus or SanDisk Industrial. Class 10 / U1 minimum. Avoid cheap no-name cards - we've seen bad flashes from Class 4 cards corrupt mid-write and add a second brick on top of the first. Price: $10-20 CAD. If you run a home lab you probably have one in a drawer.

Download balenaEtcher while you wait for the image. [etcher.balena.io](https://etcher.balena.io/) - works on Windows, macOS, and Linux, writes `.img` files to SD cards with verification. Free and open-source. If you prefer CLI, `dd if=burn-<model>.img of=/dev/sdX bs=4M status=progress` on Linux works perfectly, but pick the wrong `/dev/sdX` and you'll nuke your laptop's drive - use Etcher unless you're confident.

Flash the `burn-<model>.img` to the microSD card with Etcher. Open Etcher, click `Flash from file`, select the image Goldshell emailed you, click `Select target`, pick your SD card (TRIPLE-CHECK - picking your laptop's SSD ends badly), click `Flash`. Etcher verifies the write automatically; a red X means a bad card - get a better one. Ejection is safe when Etcher says "Flash Complete." Expected time: 2-5 minutes for an 8 GB image.

Insert the SD card into the Goldshell's microSD slot on the control board. The slot is typically on the back or side of the BOX, labeled `SD`. Some models (early KD-BOX, HS-BOX) require opening the case with 4 Phillips screws to expose the slot. Mini-DOGE's slot is under a rubber flap on the side. Be sure the card is fully seated - spring-loaded on most models, push-push on a few. If the slot is push-push and the card doesn't click, you've got it backwards.

Power-cycle with the SD card inserted. Unplug 12V. Wait 15 seconds. Re-plug. DO NOT plug Ethernet yet. The boot will now load from the SD card instead of the eMMC. You should see LED activity within 30-60 seconds that differs from the stuck red+green state - typically a slow blink pattern as the recovery script rewrites the eMMC from the SD image. This can take 5-15 minutes depending on model and image size. Do not interrupt. Do not unplug. Just wait and watch.

Watch for the "flash complete" LED pattern. When the eMMC reflash is complete, most BOX models will flash both LEDs rapidly for a few seconds, then power down the system entirely (LEDs go dark even though 12V is still applied) OR reboot to the normal green-solid state. If LEDs go dark / device powers down: unplug 12V, remove the SD card, re-plug 12V - the miner boots from the newly-flashed eMMC on the next cold boot. If the device reboots on its own: it likely rebooted into the SD image again, which means the eMMC flash was incomplete - let it run another cycle.

Remove the SD card. Cold-boot. Verify `find.goldshell.com` sees the device. Unplug 12V, physically remove the SD card, re-plug 12V, wait 2 minutes. Run the `find.goldshell.com` discovery tool. Expected: the miner appears with a new IP assigned by your DHCP. Log in at `http://<IP>` with `admin` / `123456789` and immediately change the default password. Check the firmware version matches what the `burn-<model>.img` shipped - if it's an older version, plan to upgrade (over Ethernet only!) once you've confirmed stable operation for 24 hours.

If SD-card reflash fails to complete after 3 attempts, pull the control board and inspect the eMMC. Remove the 4-8 Phillips screws holding the top cover. Disconnect the hashboard ribbon cables (note orientation - mark with a Sharpie). Remove the 4 screws holding the control board to the chassis. Inspect the eMMC chip (usually a BGA part, labelled with Samsung / SanDisk / Kingston branding). If there's visible damage - scorch marks, cracked epoxy - the eMMC itself is dead. Skip to Tier 4 (control-board swap).

Read the raw eMMC content with a USB-C reader / JTAG harness if you have one. Dump the eMMC to a file on a Linux workstation. `dd if=/dev/sdX of=goldshell-emmc-dump.img bs=4M`. Mount the partitions loop-style and inspect `/etc/`, `/usr/local/bin/`, `/tmp/`. Look for tampering - modified `rcS`, rogue cron entries, unexpected binaries in `/usr/local/bin/` or `/tmp/`. This is also where you confirm a malware diagnosis if Step 6 flagged the device.

Write the `burn-<model>.img` directly to the eMMC with `dd` via JTAG / USB-C adapter. `dd if=burn-<model>.img of=/dev/sdX bs=4M status=progress conv=fsync`. This bypasses the SD-card bootloader path entirely - use this when the SD method fails because the bootloader itself is corrupted, not just the rootfs. Verify with `sha256sum /dev/sdX` vs `sha256sum burn-<model>.img` after the write completes. Mismatch = bad eMMC block, move to Tier 4.

Verify SSH access, change default credentials, harden the device before it touches your LAN. First boot after eMMC reflash: `ssh admin@<IP>` - password `123456789`. Immediately: `passwd` to change. Edit `/etc/config/network` if you need a static IP. Disable WAN-facing services: `/etc/init.d/dropbear stop` and remove any port-forward rules on your router. Set up Wireguard or Tailscale if you need remote access - don't port-forward Goldshell to the internet. Ever.

Run a 24-hour burn-in at nameplate hashrate before declaring the repair successful. Connect to your pool (Dxpool, F2pool, or Nicehash where compatible). Monitor hashrate and temperature for 24 hours. A successful reflash should produce nameplate hashrate within 5% (e.g., KD-BOX Pro: 2.6 TH/s; Mini-DOGE: 185 MH/s). If hashrate is low or unstable after 24 hours, the eMMC is likely worn and a full control-board replacement is the real fix.

When to stop DIY. You've reached Tier 4 if: (a) 3+ SD-card reflash attempts fail to complete with 3 different quality microSD cards, (b) the eMMC shows visible damage (scorched, cracked), (c) the control board has other failures (no power LED on control, dead Ethernet PHY, fan connector dead), or (d) post-reflash hashrate is below 70% of nameplate after 24 h burn-in. At this point the eMMC or a downstream component is physically damaged - reflash can't fix hardware.

What D-Central does at the bench. We run the same SD-card recovery protocol with bench-validated 16 GB industrial microSD cards and a known-good `burn-<model>.img` library (KD series, HS series, LT series, CK series, Mini-DOGE family, ST-BOX, KA-BOX, AL-BOX II). If that fails we swap the eMMC chip (BGA rework at 220 °C preheat, 245 °C reflow) using a pre-programmed part, replace damaged Ethernet PHY ICs where the LED pattern points there, and burn-in for 24 hours before shipping back. We're the only Canadian shop documenting Goldshell-specific repair - no Zeus BTC China-ship nonsense, no 6-week turnaround, no coin-flip on whether you get your miner back.

Ship safely. ESD bag around the control board if you're sending the control only; otherwise a foam-cradled rigid box for the whole miner. Include a note: model, serial, firmware version that was being flashed when the brick happened, whether the device was ever port-forwarded, whether you attempted Tier 2 / 3 yourself (we'll know from the SD-card slot wear pattern). Ship to D-Central HQ; typical turnaround is 5-10 business days from receipt for Goldshell repairs given the `burn-*.img` email loop with Goldshell support.