EasyNT: may not be an option on that system. see slide 13 of cdrdv2-public.intel.com/783328/Inte…%20Processor%2030-3-30%20Public.pdf which I believe is for the CPU in the R260

the lanes that terminate in the CPU can either be 1x16+1x4 or 2x8+1x4. doesn't say what the chipset can do.

which one was that?

with a little digging, the PEX8800xx does work

Looks like Intel Xeon E-2400 with "C260 series" chipset.

EasyNT: if the one you're looking at is similar to that one it should see the NVMe disks

just there is a on-board enclosure device that you probably won't be able to see

what does anyone happen to know about ndi flavors?

(also, I wonder how well `NDI_FLAVOR_VANILLA` translates conceptually)

it basically does what I want, but the fasttrack and uses strongly suggest it's really an SCSA specific hack because they can't get out of abusing the driver private data

sommerfeld: Thanks so much for your research! Yes, I was thinking of the E-2468 cpu…

jbk: Thanks, where did you find that about the PEX8800xx?

i was able to find someone with the hardware and they were nice enough to boot an illumos iso and check if it could see the NMVe devices

jbk: Wow, so great. May I ask what hardware that was exactly?

i think it was a PEX880048

But do you know the card/manufacturer also?

i think it's just from broadcom

Ok I’ll check, thanks!

the mgmt device looks like it has a pci id of 1000:00b2

on linux it uses the mpt3sas driver

and looking at it... it _might_ (big emphasis) only require a few bits for our mptsas driver to recognize it

on linux basically for that device id, it's setting 'is_gen35_ioc'

and our driver has a similar setting

But you don’t know if the PEX8747 also works?

no

andyf pmooney looks good with the archive applied!

sjorge: great. thanks for the report

Does anyone here have an i40e on OI or OmniOS? If so, please utter:

ls -lt /devices/pseudo/clone@0:i40e

and share the output here?

danmcd: $ ls -lt /devices/pseudo/clone@0:i40e

crw-rw-rw- 1 root sys 1, 90 Feb 27 09:35 /devices/pseudo/clone@0:i40e

Thanks.

I have, but it’s the cursed one.

crw-rw-rw- 1 root sys 1, 86 Feb 27 19:33 /devices/pseudo/clone@0:i40e

Oh yeah, the 13230 one... so is the one I have on Kebecloud.

fenix illumos#13230

BUG 13230: i40e has duplicate traffic when used with bhyve/snoop running (New)

↳ illumos.org/issues/13230

I've noticed X722 no longer shows up on mobos. :shrug:

Pretty sure it’s a hated part

isn't "duplicate traffic when promiscuous" a well known symptom of something?

I swear I've heard of that with other parts

richlowe: If you look at kebe.com/~danmcd/webrevs/13230-newtry turns out i40e's X722 implementation doesn't honor what works on X710. The Linux i40e driver comment (repeated in my webrev) might suggest inside knowledge from Intel that I (or rmustacc) could not extract?

Testing this across all i40e parts is a nightmare, and why it's remained in stasis.

I'd done an X722 and *one* X710 part a while ago without obvious failures, but there's all sorts of corner cases too (mcast, among others).

mine was on a vendor 0x1022 device 0x15ff "X710 for 10GBASE-T" which so far has been reasonably well behaved.

richlowe: there have been other cases of duplicate traffic in the past, IIRC one of them was reported a long time ago by Theo and I addressed that. There were also issues with aggr that I believe I fixed. This case with the 700-series Intel NICs is more about minor discrepancies in how the various parts want to be programmed.

so a common symptom, not a common cause, got it

Yea the machinery around the promisc path in mac can be a bit tricky and it's also easy to mess up how/when you program promisc mode on a given NIC and how that relates to the mac groups logic (assuming the driver implements mac group support)

In the X722 case it's a problem with different 700 parts wanting slightly different variations on the programming side

sjorge and andyf ==> just approved fenix illumos#17221

BUG 17221: bhyve emulated e1000g fails with TSO (In Progress)

↳ illumos.org/issues/17221 | code.illumos.org/c/illumos-gate/+/4099

yea.... rzezeski only way to solve that one.... needs a little zig magic.

C is so 1970s

Linux shouldn't be the only project having multi-lang drama in the mailing list :P

sommerfeld: X710 has never been problematic. And oh damn, that's right, `curly` is not 722, but 710.

There are in-MNX-location machines with X722. I swapped them and curly in my head.

Is that something that could be put behind a tunable?

Not ideal but that might mean it can safely go in without testing all flavors under the sun