The SC846 SOM EZ-KIT can be unreliable at boot. Same images, and one attempt gets all the way to a Linux login while the next one just hangs partway. There's no consistent spot: we've had it die cold at JTAG (TAP comes up dead, 0 of 5 targets examine), early in the kernel around mm/SWIOTLB init, at the switch to the NFS root, and also after the login prompt where the board goes deaf to ping/ssh/serial.
Not possible to say if it's the board or the image. Take a fitImage that just hung, boot it again, and it comes up fine and passes a full-memory stress test. The NFS server end is never the problem either, mounting the export from the host is instant every time.
To get it back you power-cycle and start the debug session over. The one that really slows us down is the cold JTAG hang, where a power-cycle alone does nothing and you have to hold the physical reset button for a few seconds first.
There's a (so far unfounded) suspicion that it is a power rail. It dies at such wildly different points, from about 0.1 s in all the way out to systemd, that a single bad code path doesn't really explain it; a voltage line that sags or glitches would take the board down wherever it happened to be. The PHY/MAC dropping out after login and the dead JTAG DP from cold both fit that. Might be worth seeing if one board rev is worse than the rest.
No repeatable stress test for it yet, it just turns up over normal boot cycles. Will try to come up with something.
The SC846 SOM EZ-KIT can be unreliable at boot. Same images, and one attempt gets all the way to a Linux login while the next one just hangs partway. There's no consistent spot: we've had it die cold at JTAG (TAP comes up dead, 0 of 5 targets examine), early in the kernel around mm/SWIOTLB init, at the switch to the NFS root, and also after the login prompt where the board goes deaf to ping/ssh/serial.
Not possible to say if it's the board or the image. Take a fitImage that just hung, boot it again, and it comes up fine and passes a full-memory stress test. The NFS server end is never the problem either, mounting the export from the host is instant every time.
To get it back you power-cycle and start the debug session over. The one that really slows us down is the cold JTAG hang, where a power-cycle alone does nothing and you have to hold the physical reset button for a few seconds first.
There's a (so far unfounded) suspicion that it is a power rail. It dies at such wildly different points, from about 0.1 s in all the way out to systemd, that a single bad code path doesn't really explain it; a voltage line that sags or glitches would take the board down wherever it happened to be. The PHY/MAC dropping out after login and the dead JTAG DP from cold both fit that. Might be worth seeing if one board rev is worse than the rest.
No repeatable stress test for it yet, it just turns up over normal boot cycles. Will try to come up with something.