Troubleshooting

A failure tree for the host, driver and runtime — not a list of random reinstall commands.

No devices detected

1. Check lspci -nn.
2. Check lsmod | grep -i akida.
3. Inspect dmesg | grep -i akida.
4. Check device nodes and permissions.
5. Only then debug the Python environment.

lspci -nn
lsmod | grep -i akida
dmesg | grep -i akida | tail -n 100
ls -l /dev/akida* /dev/akd1500_* 2>/dev/null
akida devices

It worked before a kernel update

The official driver README explicitly states that the install script must be run again after a kernel update because the module is built against the kernel.

cd ~/path/to/akida_dw_edma
sudo ./install.sh

Large model / external memory failures

For AKD1500 over PCIe, CMA can become relevant when programming larger models or using more of the pipeline. Confirm kernel support and capture the model's external-memory requirement from model.summary() before treating this as a random Python error.

Device busy

Close other processes that may hold the hardware device and check who has the node open.

sudo lsof /dev/akida* /dev/akd1500_* 2>/dev/null || true
sudo fuser -v /dev/akida* /dev/akd1500_* 2>/dev/null || true

What a useful bug report contains

• Pi model and exact PCIe adapter/switch topology
• OS image + kernel
• driver commit
• MetaTF / akida version
• device-node listing
• full mapping summary if a model is involved
• exact command and full error output