Re: Radxa Orion O6

[Jared McNeill <jmcneill%invisible.ca@localhost>]

My board arrived earlier this week (thanks!) and I've managed to fix some 
of the issues you raised.


1) EFI runtime services issue

This one is still outstanding. I still feel that it's a firmware bug (that 
NetBSD doesn't properly contain, but this is a separate issue). For now 
I've committed a workaround that detects the Orion O6 board and skips EFI 
runtime initialization:

https://mail-index.netbsd.org/source-changes/2025/01/30/msg155361.html


2) XHCI crash at boot

It seems that the SoC's XHCI IP requires 32-bit aligned accesses. The XHCI 
spec defines a handful of 8- and 16-bit registers, and some of them (like 
the one that returns the XHCI version) is at an offset that is not 32-bit 
aligned. I committed a workaround for this:

https://mail-index.netbsd.org/source-changes/2025/01/30/msg155355.html

Once I fixed this, the crash went away, but USB devices wouldn't attach 
("device problem, disabling port n"). I tracked down the root cause for 
this and committed a fix as well:

https://mail-index.netbsd.org/source-changes/2025/01/30/msg155360.html


3) Crash when booting with SMP enabled.

The crash was caused by virtual timer interrupts not firing on CPUs 1-4. 
The first workaround I tried was to use the physical timer -- this worked, 
but not ideal. So I did some more digging and it turns out NetBSD was 
making assumptions about the state of the CNTHCTL_EL2 and CNTKCTL_EL1 
system registers when entering the kernel that were not valid. I fixed 
initialization for both of these:

https://mail-index.netbsd.org/source-changes/2025/01/30/msg155358.html
https://mail-index.netbsd.org/source-changes/2025/01/30/msg155359.html


If you update your tree to pick up all of these changes, you should be 
able to boot an unmodified kernel to the installer.

Take care,
Jared