Looks good.

I was able to build kboot for powerpc64, with diffs D33511 - D33515.
In the machines that I have access to, Petitboot can't run BE loader.kboot, but I was able to build and run a LE loader.kboot.
However, as mentioned in D33511, it can just run some loader commands, but can't list files or load a kernel. But the older version, without these changes, also can't.

Works fine, after the proposed fix in D33511.

Is this change good to go now?

Nice!

I've tested these changes on a PPC64 VM. With them, the virtual SCSI device fails to attach, making it impossible to mount root.
Below are the relevant dmesg messages:

Build uboot for ppc32 too

Initialize RGB offsets along with palette.

Simplify sysctl setup

@scottl do you think this is ok now?

Address reviewer's comments

Remove obsolete endian conversions

Initialize archsw earlier.

Address reviewer's comments

Add missing break

Query device tree for Radix/GTSE support, instead of using CPU version and HV bit checks, as this is a better approach, as discussed in IRC.

• Address reviewer's comments

Address reviewer's comments:

• log DMA load buffer error message only once for each device
• add a per device error counter, exported via sysctl (inspired by mpr code)

Worked fine on POWER9.
For PowerPC, without these changes the allocation of more programmable counters than available already failed, because of the check for the counter number in allocate. But, as you said, it is always a good idea to validate the PMC class too.

In D31109#704608, @mhorne wrote:

Looks good, thanks for finishing this. Will you handle the removal of the static power8 definitions as well?

This approach of auto detecting the correct frame buffer physical address based on Vendor ID looks good.
It would help to make FreeBSD graphics work out of the box for Blackbird and Talos II machines, instead of forcing the user to figure out the physical address of its VGA adapter and pass it to the kernel manually, or else the system just hangs.
Ideally, Petitboot should fix it on their side, but I don't know if it will happen anytime soon, since Linux isn't affected, because it just talks directly to ASPEED video device, instead of relying on the framebuffer info exposed in the device tree.

D31232 fixes OFWFB with radix. After it lands it should be safe to make radix default.

@imp I've added the rate limiter, do you think this revision is ok now?

• Fix event name mapping
• Handle non-programmable PMCs (PMC5 and PMC6)
• Add missing event counter check

Commandeering revision to finish changes while Leonardo is out.

LGTM. The changes are similar to the PPC64 ones, in D31109.

I have investigated the issue further and have some suggestions of how it can be fixed (inline).

Overall the changes look good.

Address review's comments

Address review's comments.