[illumos-gate] 17179 drv_ioc_prop_common() needs some errno improvements -- Patrick Mooney <pmooney⊙pc>

Hi everybody, the HCL at illumos.org/hcl gives a 504 Gateway Time-out, has anybody contact to the right person to fix that?

same here

EasyNT: it is known the person is informed

sjorge: you around?

toasterson: Thanks a lot!

andyf I am not, but if you shoot me an archive I can test them this weekend

pmooney I am now but not for super long

andyf I'm currently running omnios-bloody-20250208 (well a BE after you did a full rebuild with the viona MTU fix in place)

sjorge: I was going to suggest a quick test of disabling TSO in the guest which was tripping over the e1000g stuff

Not sure I can

(probably quicker than applying a hotpatch)

rgr

It's a windows 11 vm

I've always used virtio

ah

But just switched it to e1000 to confirm the issue someone was having on the mailing list

Windows basically never makes it to the login screen

aye

I'm assign it trips immediately after it loads the driver

I see andy's sync has "bhyve e1000: Sanitize transmit ring indices." could be interesting to test again what that applied

wait... nvm I was looking at an old sync notes from 2023

at least it's repeatable, I just tried again

But the crash seems to also have cause some corruption on disk

windows now tries to boot in recovery mode before crashing

same stack

let me quickly try with MTU 1500 before I go back to work

same

AHA!

Win 11 triggers it, Ubuntu LTS is fine, FreeBSD is fine

andyf I'm guessing you did not try to replicate it with windows 11?

anyway, back to work I go now

fenix illumos#17220

BUG 17220: rtld: build with NO_SIMD (In Progress)

↳ illumos.org/issues/17220 | code.illumos.org/c/illumos-gate/+/4098

Also that means it could have been broken for a long time, I used e1000 when the MTU >1500 temporarily broke things for me. But that was on a linux VM.

So sadly that means it's hard to pinpoint when

sjorge: I think we have a good understanding of the bug (thanks to pmooney). I have a fix out for review and I'll drop you a hotfix later today so you can give it a spin over the weekend. Thanks for the help.

Does anybody know if the Broadcom PEX8747 chipset is supported under Illumos? Specifically this card here: delock.com/produkt/90078/merkmale.html?f=s

Unsure. If it presents like a PCIe bridge it shouldn't be a problem, BUT you never know so I can't say with any certainty, unfortunately.

(I thought NVMe expanders were PCIe bridges with M.2 slots. Happy to be corrected/educated-further.)

As far as I know the bridge thingies are just one M.2 slot per PCIe slot, unless they do what’s called Bifurcation. This card has it’s own controller, the Broadcom PEX8747, which among other sutff has S.M.A.R.T and boot support. Hence the PEX8747 should be supported by the OS, otherwise money for nothing (and still no chicks for free) :D

the product brief for that chip says that the software model is that of pci-pci bridges, so i'd guess it might just work

linux-hardware.org/?id=pci:10b5-8747-10b5-8747 reports it as PCI switch

You could go with Framework. It's 20cm case with 128GB RAM

You're talking to me @toasterson ? Good to know. If you wanna see the history of my HDCs (for reference), see these posts: kebe.com/blog/?tag=hdc

I'm starting to gather wanted-improvements, and while some of them could be addressed with HDC 3 improvements, some of them can't.

Honestly, apart from the case's lack of 3.5" HDD slots (which isn't a HUGE problem), HDC 3 is holding up better at 4.5years in production than either of its two predecessors.

danmcdI used it for reference for my last one :) But now I am on a AMDGPU with my current one. So OpenIndiana is now inside Vagrant boxes for testing

Good evening everyone, does anyone know why the page illumos.org/hcl hasn't been accessible for a few days?

web and crawler issues

if that card is the one i'm thinking of, it should just look like a child PCI bus, though IIRC I _think_ it also presents a SAS HBA in order to present some SES bits for the LEDs it can control (if the actual slot for the NVMe has any lights)...

something like that...

a pci bridge that's also a SAS hba?

like... at the same time?

something like that

like on linux, they want you to also install their sas hba driver

even though the individual NVMe devices show up just fine

I don't think it's trying to do any sort of raid or such (thank god)

but I guess just since there's no real PCI(e) equivalent of an SES device

and it's i think the same driver used with their other HBAs

(and you see the devices even w/o)

EasyNT/danmcd: generally simpler and cheaper (if the motherboard supports quadfurcation) to get one of the many inexpensive 4x M.2 cards that just pass the PCIe lanes through and don't require a bridge.

Woodstock & toasterson: Great news, thanks for pointing that out!

sommerfeld: But that would then result in less throughput per NVMe, wouldn’t it?

EasyNT: why would it? there are still the same number of PCIe lanes driving each M.2 slot. when you enable bi/quadfurcation, the chipset treats the slot like the 2 or 4 independent PCIe devices it actually is.

bi/quadfurcation: Thanks for clarification… So then I guess I’m left with the question wheather a Dell PowerEdge R260 is capable of doing a quadfurcation…

sommerfeld: Sorry, had the wrong stuff in the clipboard, I meant to mention you :)

EasyNT: and that's a question that would most likely be answered in a traversal of the BIOS setup menus. My hands-on experience with PCIe quadfurcation is limited to a single system using a Tyan mobo and an AMD CPU.

sommerfeld: Yes, I’m trying to figure that out with Dell manuals, since I don’t have the machine yet and am trying to figure out if I should buy that thing :).

the impression I get (and I may be totally wrong about this) is that bi/quadfurcation is more likely to be present on recent AMD-based systems (vs Intel).

sjorge - bhyve e1000g hotfix is at hf.omnios.org/bloody/bhyve-e1000g.p5p

andyf: did you do a tar one for nomad?

I'm pretty sure I did, a while back.

ok, I think I'm confused about things then

But perhaps not, since I can't find one..

Ah no, it was jclulow who built a binary for nomad - log.omnios.org/omnios/2025-01-02#1735859095-914893

jmclulow did it, according to the ticket. (17015)

right, no m in his nick. I don't know why I always want to put it there.

there _is_ an m there

but not in the nick

Either way, I'm looking forward to it rolling out in OmniOS :)

sommerfeld: The manuals do not mention it, but bigger Dell systems have it. I don’t know if this is just a software issue (BIOS not supporting it) or if there must be some actual physical support on the board… So I probably still have to go with the Broadcom PEX8747 bridge…