anyone ever see / experience [ID 433237 kern.warning] WARNING: t4nex1: regs[2] does not have a valid MMIO address

i have a bunch of compute nodes with chelsio T62100-lp-cr cards

Sometimes the CN some up clean with 2 and now its not. the difference between the ones that boot normal and one that doesnt can be seen here. gist.github.com/Smithx10/0ece643e3f85207390220c62a1fad930

Smithx10: I've not seen that before. That'd suggest something in firmware or later pci_boot.c is ont programming that correctly.

I would grab the prtconf output from it.

rmustacc: danmcd suggests something like a partially plugged in card. prtrconf output gist.github.com/Smithx10/8fc260abf19e972634d1ca9759336f04

Smithx10: You need -v to get properties.

Looks like after re-seating both the risers and cards in one of those boxes we have link lights on both cards.... its booting now, see if we still have the error

yeah that server came up clean afterward. Im assuming the risers werent seated fully

OK. Doesn't quite add up for me, but fair enough.

alright, I just gave the order for him to reseat the 4 others that were effected out of the 20

if any have the same issue ill get a -v sent over

But I'm also having trouble tracking down exactly where that error is being geerated. Ultimately if it is that and that solves that, all the better.

@rmustacc i got a prtconfv dump from one us-east-storage.solutions.iqvia.com/bruce_dev/public/prtconfv

but I think that one isnt even showing the other card... looks like it might be the seating

hrm...

we have errno.d that provides sympolic values for error values (i.e. `inline int EPERM = 1;`)... any feelings about adding a script that can be #included that maps values back to name (in the same vein as truss)?

I think you probably just want a subroutine or action that does this ala strerrorname_np() but without the mouthful.

At least, that's what I had in the back of my head as a follow up.

This might be a long shot, but does anyone have a good source to buy aftermarket parts for SPARC hardware outside of eBay? Specially for an M10-4...

Smithx10: the message usually means we've been given an MMIO address of 0x0, or near it

not admin error, bug.

rmustacc: if y'all are trying to make libjedec public, I'd expect an IPD?

if only to say so, and maybe update some manuals

richlowe: I'm not intending to make libjedec a committed interface at this time.

so it's a kind of contracted private affair?

if that's the case, I think an ipd makes even more sense to warn people, but also we've never done that before since the Sun days

and it didn't entirely work great even then

Yeah, I don't think we'd be making any promises like that all imho.

As I mentioned in the ticket, it may makes sense in a 100% buyer beware case like we do for other libraries for folks to experiment with. But reserving all rights with prejudice I guess.