rmustacc: are there any examples of enumerating a locate LED in fmd for a specific type of card?

there's kinda something for nics, but those are ports, not the card

jbk: What provides the facility?

right now nothing.. that's I guess I'm trying to figure out.. the card (third party driver unfortunately) has an IOCTL to turn an ident light on/off.. it'd be nice to be able to hook that into fmd, but I can't see any obvious way to trigger a module to enumerate anything based off the driver

on a somewhat related note, any thoughts on maybe adding a flag to 'dladm show-phys' (seems most logical place) that could use the fm topo stuff to list a location for a NIC (somewhat analogous do diskadm -P)?

[illumos-gate] 15715 rdsv3: the comparison will always evaluate as 'true' -- Toomas Soome <tsoome⊙mc>

e.g. I was just dealing with a system w/ 8 instances of mlxcx

where (for reasons) the instance ordering changed

but the PCI slots were labeled, so I could at least use fmtopo to map by hand mlxcx0 -> PCI slot an such

does prtdiag show it?

granted for onboard NICs it's probably less useful w/o a topo map since most would just show up as 'MB' (though even there could help if there's a mix of onboard/cards)

prtdiag shows the slots and if they're occupied

but nothing obvious to tell you how that maps to drivers & instances

yep, oracle also does only this:

ID Status Type Description

--- --------- -------------------- ------------------------

1 in use PCI Express Gen3 /SYS/MB/RISER1/PCIE1

2 in use PCI Express Gen3 /SYS/MB/RISER2/PCIE2

3 available PCI Express Gen3 /SYS/MB/RISER3/PCIE3

Is it a third party driver you can modify?

yes

So we already do device-class specific logic in the pci enumerator to go off and do things.

So the ways to approach this vary.

Because it's tied to a card and not something else, I presume that we don't want to do something like use static nodes.

That's for example, what I've done for our drive bays and other things.

yeah the card could be in any slot

The topo pci enumerator does allow for PCI class-specific things to happen, but those are also not driver specific or it'd be a bit weird for the driver here.

The simplest path would be to add a custom facility and just enumerate it in the PCI enumerator locally.

If we wanted to avoid encoding that in the PCI enumerator, there are probably some other ways to approach it and add something like ksensors.

But that's a lot of work and I'm not sure if that'd help us a ton.

What's the hardware out of curiosity? Not a datalink by chance?

no

HBA

I missed stuff above, but oracle's CRO stuff jmcp did is lovely.

CRO?

"Chassis, Receptacle, and Occupants"

ahh.. so not related to mst3k :P

not officially

but then officially, it's just a coincidence the linker aliens created commands named lari, crle, and moe, and not related to any stooges

i do like the idea of devchassisfs (linux kinda does something vaguely similar)

jbk: I think they did it initially because of the large OPL, where your princess might be in a whole other rack

but even on smaller machines there's been times I would have a much happier and more direct experience if I could refer to things based on where they were (and refer to where they were based on the things that were there)

just outside of OPL, almost always with disks