IceRiver SD Card Reflash Recovery Procedure

Hard power-cycle for 60 seconds at the breaker. Full mains kill — not a soft reset, not a power-button press. About 10% of 'needs SD recovery' tickets we see actually resolve at this step alone because the failure was a wedged daemon state surviving soft reboots. Always try this before burning an SD card.

Hold the reset button for the full `20 seconds` until the red status LED starts flashing, then release. Wait `5 minutes` untouched. Factory reset on KS hardware copies the factory partition to the active partition. If your problem was config-level rather than rootfs-level this recovers it without SD recovery. If nothing visible happens within 5 min, the active partition was already too corrupt — proceed to SD recovery.

Verify network path before assuming you need SD recovery. Power on, wait 5 min, check your router's DHCP lease table for the unit's MAC. If the unit has a lease and answers `ping` but port `80` is dead, you're a candidate for SD recovery. If the MAC isn't in the lease table at all, swap Ethernet cable and switch port first — dead Ethernet ports masquerade as bricks.

Catalogue the trigger before reflashing: was this mid-OTA over Wi-Fi, mid-OTA over Ethernet, after `xyys`/`tswift`/`iceriver-oc` flash, after a power event, what firmware version was running before? This tells you whether to expect clean SD recovery or expect Tier 3 escalation, and saves D-Central 30+ minutes if you eventually ship the unit.

Try IceRiver's 'Detect IP' tool from `iceriver.io`. Some half-bricked units present as 'no DHCP' but are actually answering on `192.168.1.1` (factory default fallback) because DHCP failed. Plug the unit directly into a laptop with a static IP on `192.168.1.0/24` and try `http://192.168.1.1`. Free, takes 2 minutes, occasionally saves the SD recovery step entirely.

Burn the recovery image to microSD with `balenaEtcher`. Download Etcher from `https://etcher.balena.io/` — free, cross-platform, validates the write after burn. Insert your tested 8-32 GB Class 10 microSD. In Etcher: select the `.img` file from `iceriver.io/firmware-download/`, select the microSD as target, click 'Flash.' Wait for both write AND verify passes to complete. Don't skip Etcher's verify pass — it catches bad SD cards before you've spent time on a failed recovery.

Power off the miner at the breaker and let it sit for `30 seconds` for full discharge. Some KS firmware revisions hold a 'boot mode' flag in non-volatile RAM for a few seconds after power-off, and an immediate SD-recovery attempt may land before the flag clears. Wait the full 30 seconds.

Insert the prepared microSD with the unit fully powered off. Locate the SD card slot — varies by KS model (KS0/KS0 Pro: slot near Ethernet jack, often externally accessible; KS3/KS3M/KS5L/KS5M: slot on the long edge of the control board, usually requires Phillips #2 to remove the chassis lid). The card should click into place. Confirm correct seating before powering on — a half-inserted card during boot can corrupt both the card and the eMMC.

Power on while holding the reset button for `5 seconds`. Power on at the breaker, immediately press and hold reset. Hold for 5 s. Release. Some KS firmware revisions want the button held longer (up to `30 s`); if 5 s doesn't trigger, retry with the longer hold. Watch front-panel LEDs — successful recovery starts with a distinct LED pattern (typically all four LEDs cycling rapidly, then `D1` solid green, `D2` flashing) within 15 seconds of release. If you see this pattern, recovery has triggered.

Do not power-cycle, touch, or reseat anything for the full `8 minutes`. SD recovery on KS hardware writes both rootfs and (sometimes) bootloader partitions — interrupting mid-write produces a deeper brick than the one you're recovering from. LEDs will cycle through a full recovery sequence over 3-8 minutes depending on model and SD card speed. The unit will reboot itself once recovery completes; do not assist.

Verify the SD card was burned correctly if recovery doesn't trigger. Pull the card, plug into your laptop, check that the partition layout matches what the IceRiver image expected (most SD recovery images create a `boot` partition with specific magic-header files). If the card looks empty or has wrong partition structure, re-burn with Etcher's verify pass. Test with a fresh image from a different source if available.

Try alternate trigger sequences if the standard 5-second hold doesn't trigger recovery: (a) hold reset for the full 30 seconds during power-on, (b) power on without holding reset, then press-and-release reset within 15 seconds, (c) for KS0/KS0 Pro specifically, try inserting the SD after the unit reaches the splash screen and then holding reset. The IceRiver firmware-download portal lists model-specific procedures; the IceRiver Discord tracks per-revision quirks.

Confirm the SD card is being read at all. Open the chassis with the unit powered on and SD inserted, watch for the SD card slot's activity LED (most KS control boards have one near the slot). If the activity LED never blinks during the 15 seconds after power-on, the bootloader isn't reading the card — the SD recovery code path isn't engaged. Escalate to UART (Tier 3 in the bricked firmware recovery page).

Get UART access and confirm bootloader behaviour. USB-to-TTL adapter (`CH340`, `CP2102`, `FTDI`, `3.3 V` logic). Connect to debug header (location varies by model). `PuTTY` at `115200 8N1`. Power on with SD inserted. U-Boot output should report SD card detection. `mmc0: cd not asserted` = bootloader not seeing card; `signature mismatch` / `bad magic` = wrong image; U-Boot prompt without SD selection = recovery path not compiled into your firmware revision (proceed to UART-only reflash per bricked firmware recovery page).

Verify firmware integrity post-flash. After successful SD recovery, run the miner for a full hour at nameplate hashrate. Check the miner log for `mmc` errors, eMMC write errors, or I/O retry warnings. A 'successful' recovery that produces I/O errors during normal operation means the eMMC has marginal blocks — the brick will return within days or weeks. Plan for control-board replacement at Tier 4.

Stop DIY and ship to D-Central when (a) SD recovery doesn't trigger across multiple cards, multiple trigger sequences, multiple known-good images; (b) UART reveals bootloader corruption (`bad header`, `mmc init failed`, no output at all); (c) recovery completes but eMMC errors return within 24 hours; (d) you see scorched components, swollen capacitors, or PMIC damage during inspection; or (e) the brick happened during a power event and downstream hardware damage is suspected.

D-Central bench process: multi-revision SD image library (we maintain canonical images for every KS hardware revision encountered), UART access with stable bench power, full eMMC dump and analysis, eMMC chip desolder + external programmer reflash for irrecoverable cases, control-board swap with image transfer when chip replacement is uneconomic, full PMIC continuity testing for power-event bricks, post-repair 24-hour burn-in at nameplate before sign-off. Western retail bench — no shipping to China, no Zeus-style trust gap.

Ship the control board (or whole unit) in anti-static bags, double-boxed with at least 5 cm foam on every side. Include a printed history: original firmware version, brick trigger, which Tier you reached, which SD images and trigger sequences you tried, which model + hardware revision (chassis sticker). This saves D-Central 30+ minutes per unit and the savings come back directly in the repair quote.