[illumos-gate] 17468 comm page should handle larger NCPU -- Patrick Mooney <pmooney⊙pc>

OK wow! I didn't know this worked so well but saw a few CR's over the past year from the stlouis branch retlatd to it

So I just marked a NVMe that was faulty by blinking the led, offlined it, replaced it, hotplug worked it showed up and now replacing the offlined device without rebooting the host!

on OmniOS, I'm trying to understand why a shared-ip lipkg zone does not seem to honor the global zone's routing table to a local subnet, and sees packets sent to the default gateway instead

after getting the word "tribblix" included in an Oracle blog post last month, this month I went with "illumos"...

aww

YES!

Has anyone booted any illumos distro on one of those Framework laptops?

I should probably try some USB[34] and or TBolt experiments on one of my NUCs before asking that. Sorry.

sjorge: That's great NVMe news!

veg: i'm guessing the shared-ip zone doesn't have an ip address on that local subnet?

any tweaks to the "hostmodel" protocol property in ipadm?

danmcd: blackdot.be/2025/06/illumos-nvme-hotplug bit rough, as I wasn’t planning on posting. But it seems to have sparked some interest.

sjorge: a810 isn't in the upstream database.

So that's why it says unknown dvice.

We'll have some seriously hot days but hopefully nothing like that.

sorry, mis-chan.

rm: yeah, I wasn’t 100% sure, it’s a relatively newish model. I’ll edit the post next time I’m at my computer

sjorge: IIRC, if the zfs fru info is populated with the physical location (zdb -l /dev/rdsk/.....s0 (or s1 if a UEFI boot partition is present).. autoreplace might even work

Well, when data is concerned I’d like some form of manual control 😅

I am wondering now if I could hotplug offline a ppt device, update ppt_* files and online it with original driver or reverse

sommerfeld: no ipadm tweaks at all, and yes, the shared-ip lipkg zone has a 10.10.10.0/24 address, while the global zone has a 10.20.20.0/24 address

packets flow alright, but the share-ip zone uses the default gateway instead of grabbing the shortest route from the GZ routing table

Fixed the blog post, I guess eventually the device will get added once we sync the... pcids file thing which seems to happen every year-ish ?

No, I mean the upstream database doesn't have a record of it.

it happens when someone remembers, pretty much.

so if the device gets added, you could sync it and submit that if you wanted

I can probably ferry device information if useful. Though it's easier when I have the datasheet to confirm.

there's no reason to wait for someone else to do it, is what I mean

upstream in this case being pci-ids.ucw.cz

Yeah. I try to add a bunch of SSDs and AMD stuff as I can.

speaking of pci things, did we ever learn what's up with jbk's -1 lanes?

I've been a bit busy. But I have everything I need to get to final bit there.

Right now the code says if there is no current width property it stores -1.

So to me that's the most practical thing.

Yeah, that's the problem. Basically this is a bridge that doesn't have a downstream device therefore there is no current link width property.

I'll get something figured out for this over the next day or two.

But there's no -1 read.

That's the answer. Thanks for the patience and data on this jbk.

Is this part of why the larger pci code is suspicious of empty bridges?

Dunno.

This is all in pcieadm.

yeah, I just remember that one of the workarounds in the pci/pcie code is to skip them (unless hotplug capable, I assume), and I didn't understand why

The main reason is that if they're not hotplug-capable then by definition nothing can ever show up in them if there isn't a device present then.

Strictly speaking if there is a device present then the link can go up and down at run-time though.

But I think basically the theory of the firmware spec and others (versus reality) is that if there's nothing detected in the slots at initial training, then no resources should be spent on the bridge if not hotplug capable.

Some BIOSes will even go back and hide the bridges from config space and acpi entirely.