That's very useful. I'm nowhere near as practiced with this, so it'll take some time to digest and plan a way forward. It does fit with what seemed to be the case, though: it's device-level issues. I'll start poking around in the specific device differences and see what shows. The great hint you've found is that this is likely affecting all CSI cameras, not just the IMX708. That's most useful. I'm not at all worried about networkmanager, because the only tweak I made in the custom build was to use systemd-networkd, which (for me, anyway) is more straightforward for embedded/minimal/CLI usage. What I haven't yet done, and probably should, is switch to debugging this in something more vanilla, like a Pi 4. At least then I can compile libcamera and friends directly on the device. The Zero 2W should be the same -- he says, hopefully.

Greetings ### [Follow-up Update] Boot Deep-Dive via Cubie-A5E DTB Masking Following up on my previous boot freeze post, I (with some assistance from Gemini - who formatted this post) managed to force the Armbian kernel to initialize by masking the sister chip profile (`sun55i-a523-cubie-a5e.dtb`) over the broken `sun55i-t527-orangepi-4a.dtb` target and injecting `clk_ignore_unused` / `irqpoll` parameters. This got us past the initial handoff hang and successfully spawned the 8-core CPU array initialization and initramfs launch (`Run /init as init process`), but the system eventually hit a permanent deadlock during the device probe phase. I don't think I can get this to go any further without some many hours spent looking at how these things work and its currently 35 degrees C here in the UK so time for a beer methinks! #### Full Boot Log Paste Link: - this is the log before using irqpoll parameter (which eventually caused a deadlock) - I can post the irqpoll logs although i wouldn't have thought irqpolling is the way to go ? https://pastebin.com/TvxTxdbp ### Either way - having just read that this post refers to irqpoll then this is the boot logfile with irqpoll added an an extraparm https://pastebin.com/TNS3DDTC #### Key Technical Takeaways from the Log Dump: 1. **PMIC / I2C Communication Failures:** The kernel initializes the AXP717 and AXP323 power chips but encounters standard timeout errors right away: ```text axp20x-i2c 0-0034: AXP20x variant AXP717 found axp20x-i2c 0-0034: Failed to set masks in 0x40: -6 axp20x-i2c 0-0034: failed to add irq chip: -6 ``` 2. **MMC / Storage Controller Deadlock:** Because the AXP717 framework drops out and the Cubie pin definitions do not line up with the Orange Pi 4A's physical board wiring, the kernel enters an endless loop waiting for internal voltage rails (`cldo3` and `bldo1`) to power up the storage controller: ```text platform 4020000.mmc: deferred probe pending: platform: wait for supplier /soc/i2c@7081400/pmic@34/regulators/cldo3 platform 4021000.mmc: deferred probe pending: platform: wait for supplier /soc/i2c@7081400/pmic@34/regulators/bldo1 ``` This prevents the storage device slot from spinning up, trapping the system right as it attempts to mount the root filesystem. Hopefully this helps isolate the unaligned power configuration loops and clock tables in the current `armbian-imager` source profile for the OPi 4A!

Kraid is a new Rust-based compiler for Panfrost that replaces the aging Bifrost-rooted compiler stack with a cleaner, more flexible design for modern Mali GPUs, improving IR structure, register allocation, hardware testing, and long-term maintainability. View the full article

Greetings - just investigating an issue with the opi-4a Armbian 26.8.0 minimal (edge 7.0.12) build - not getting past starting kernel. The board has a T527 - Hopefully this information is useful ? ### [Bug Report] Orange Pi 4A - Silent Hang at 'Starting kernel ...' (Trixie Minimal) **Board:** Orange Pi 4A (Allwinner A523/T527) **Image used:** Debian Trixie Minimal (Built via armbian-imager) **Installation Media:** SD Card (Class 10) **Log Type:** UART Serial Console Output (115200 baud) --- #### Description of Issue The recent community build of Debian Trixie Minimal for the Orange Pi 4A successfully clears the U-Boot / SPL stages, loads the Ramdisk and DTB from the SD card, but hangs completely silently immediately after hitting `Starting kernel ...`. The onboard status LED remains solid red with no flashing activity. #### The Technical Root Cause Looking closely at the UART initialization text, there appears to be a mismatch between the SoC detected by the low-level firmware and the Device Tree Blob (DTB) loaded by the Armbian boot script: 1. **Firmware/U-Boot Chip Detection:** TF-A and U-Boot explicitly identify the SoC as an **Allwinner A523**: ```text NOTICE: BL31: Detected Allwinner A523 SoC (1890) CPU: Allwinner A523 (SUN55I) ``` 2. **Script DTB Selection:** Immediately afterward, the automated script loads a **T527 DTB** profile: ```text Load fdt: /dtb/allwinner/sun55i-t527-orangepi-4a.dtb ``` Because the kernel tries to initialize the primary power rails and clock gates using a T527 map on an A523 layout, the CPU locks up silently at the execution handoff. --- #### Complete Boot Log (Up to the Hang) ```text U-Boot SPL 2026.01_armbian-2026.01-S127a-P60fe-Hc6a9-Vf757-Bd0d2-R448a (May 30 2026 - 05:53:33 +0000) DRAM: 2048 MiB Trying to boot from MMC1 NOTICE: BL31: v2.13.0(debug):armbian NOTICE: BL31: Built : 05:52:43, May 30 2026 NOTICE: BL31: Detected Allwinner A523 SoC (1890) NOTICE: BL31: Found U-Boot DTB at 0xa0b8ac0, model: OrangePi 4A INFO: Configuring SPC Controller INFO: BL31: Platform setup done INFO: GICv3 without legacy support detected. INFO: ARM GICv3 driver initialized in EL3 INFO: Maximum SPI INTID supported: 287 INFO: BL31: Initializing runtime services INFO: BL31: cortex_a55: CPU workaround for erratum 1530923 was applied INFO: PSCI: Suspend is unavailable INFO: BL31: Preparing for EL3 exit to normal world INFO: Entry point address = 0x4a000000 INFO: SPSR = 0x3c9 U-Boot 2026.01_armbian-2026.01-S127a-P60fe-Hc6a9-Vf757-Bd0d2-R448a (May 30 2026 - 05:53:33 +0000) Allwinner Technology CPU: Allwinner A523 (SUN55I) Model: OrangePi 4A DRAM: 2 GiB sunxi_set_gate: (CLK#35) unhandled Core: 94 devices, 24 uclasses, devicetree: separate WDT: Not starting watchdog@2050000 MMC: mmc@4020000: 0, mmc@4021000: 2, mmc@4022000: 1 Loading Environment from FAT... Unable to read "uboot.env" from mmc0:1... In: serial@2500000 Out: serial@2500000 Err: serial@2500000 Net: No ethernet found. starting USB... USB EHCI 1.00 USB OHCI 1.0 USB EHCI 1.00 USB OHCI 1.0 Bus usb@4101000: 1 USB Device(s) found Bus usb@4101400: 1 USB Device(s) found Bus usb@4200000: 2 USB Device(s) found Bus usb@4200400: 1 USB Device(s) found scanning usb for storage devices... 0 Storage Device(s) found Hit any key to stop autoboot: 0 PCIe: Acquiring resources... PCIe: Failed to get switch-sel GPIO: -2 PCIe: Hardware init failed with error -2 Device 0: unknown device switch to partitions #0, OK mmc0 is current device Scanning mmc 0:1... Found U-Boot script /boot.scr 4641 bytes read in 2 ms (2.2 MiB/s) ## Executing script at 4fc00000 U-boot loaded from SD Boot script loaded from mmc 249 bytes read in 2 ms (121.1 KiB/s) Load fdt: /dtb/allwinner/sun55i-t527-orangepi-4a.dtb 33383 bytes read in 7 ms (4.5 MiB/s) Working FDT set to 4fa00000 Failed to load '/dtb/allwinner/overlay/sun55i-t527-fixup.scr' 16748327 bytes read in 814 ms (19.6 MiB/s) 41171456 bytes read in 1707 ms (23 MiB/s) Moving Image from 0x40080000 to 0x40200000, end=0x42a30000 ## Loading init Ramdisk from Legacy Image at 4ff00000 ... Image Name: uInitrd Image Type: AArch64 Linux RAMDisk Image (gzip compressed) Data Size: 16748263 Bytes = 16 MiB Load Address: 00000000 Entry Point: 00000000 Verifying Checksum ... OK ## Flattened Device Tree blob at 4fa00000 Booting using the fdt blob at 0x4fa00000 Working FDT set to 4fa00000 Loading Ramdisk to 49007000, end 49fffee7 ... OK Loading Device Tree to 0000000048f96000, end 0000000049006fff ... OK Working FDT set to 48f96000 Starting kernel ... ```

NEWS! I got Mali450 on kodi! I compiled my kodi version with the dockerfile attached. About the acceleration, quoting Claude: EDIT: to make KODI working just set Settings → Player → Videos → Render method → Direct to plane Dockerfile.cross