[illumos-gate] 16988 cxgbe should properly define and enumerate -- Patrick Mooney <pmooney⊙pc>

a bit of a random question, but anyone know if anyone's making systems like pc engines did (small appliance-ish form factor), but with 10gb NICs?

[illumos-gate] 17299 libpool: dprintf() cleanup -- Hans Rosenfeld <rosenfeld⊙gho>

jbk: eu.protectli.com/vault-6-port

not quite as small as the pc engines stuff, though

but not too bad..

i use the 4-port model with 2.5g nics as a home router

managed to squeeze a 2nd m2 nvme in there using an adapter for the wifi port :)

yeah, i'm getting fiber here in a bit... though initially just doing 1gb, but they currently offer up to 5gb

the builtin mmc flash is currently not supported by illumos, something that I intend to fix if I can find the time for it

we do have an mmc driver, right?

other than it "just works", i run smartos on mine with a netbsd vm as a router/firewall

we have an sdhost driver, which apparently is the same thing

it's just hasn't been updated in a long time, even the sd card slot on my 11 year old thinkpad isn't working with it :)

ISTR seeing it when adding the plumbing for TRIM/UNMAP/... ops to blockdev

yes, that was one of the early uses of blkdev

yeah

i can't remember if i actually added it (I'm thinking no because I had no way to test) or just looked to see how it was done

since there was heartburn over blkdev supporting multiple ranges to free/trim/... in a single request

jbk: I'm in a similar situation (10g fiber buildout happening in my neighborhood - they're claiming install in June) and I'm very close to buying one of those protectli 6-port units.

if i really wanted to get fancy, there's a 2nd provider (the incredibly generic sounding 'consolidated communications') offering gb fiber as well, but the consensus is that their service is terrible

vs the local power co-op offering fiber in their service area

I read «10g fiber» as 10 grams of dietary fibre

and I wondered why a network infra buildout would need dietary fibre

MelMalik: otherwise the bits get stuck in the pipe...

that seems opposite of how pipes work, you must understand that these are not human guts

if you put dietary fibre down a sink more often than not it will just clog

.... you would it take over 30,000 bowls...

??

an 'old' SNL commercial

'colon blow'

spoofing ceral commercials at the time that kept trying to 1 up each other in how much fiber they have

good lord.

sorry for the distraction

Anyhow sonic.com is doing a buildout in my neighborhood at the moment (seeing their trucks on a regular basis). They're offering 10gbits/s symmetric for $50/month.

nice

they're charging $99/mo for gb here, but only $149/mo for 5gb

though that's still less than half the cost of comcast

err excuse me.. xfinity

the 6-port protectli has one of the intel cpus with big/little cores (2 big, 4-6 little IIRC) which prompted my question about asymmetric cores last week.

(or whenever it was)

ISTR that intel didn't match CPU features vs. just speed differences, which makes scheduling 'interesting'

at least in some cases...

jbk: my understanding is that Intel has "fixed" that (by disabling AVX-512 on the big cores) but it took them a few tries before they disabled it in a way that system vendors couldn't evade...

jbk: seen some indication that there's some other minor detectable behavioral differences between cores but largely in the realm of stuff that is unspecified in the architecture (value of the overflow flag after certain instructions is unspecified)

intel's high-level guidance for scheduling seems to be to put one thread on each P-core, then fill the E-core's, then put a 2nd thread on each P core (E cores aren't hyperthread).

man...

what a waste of the avx-512 capability

Woodstock: I recall poking for a while at the sdhost stuff for eMMC on a Wyse 3040 and ... I did not get there haha. Would be happy to look at and test changes if you end up making them!

(I ended up just booting those machines from a ramdisk)

that is... a ramdisk image on a bootable USB thing

i borrowed a 2nd protectli vault from a friend just for this purpose, now I just need to find the time :)

MelMalik: eh, market segmentation. it was cheaper for them to stamp a known working core on the die than to validate a slightly smaller one with avx512 taken out. Happens all the time. (first one I remember: HP slowed down the FPU in the 9000/712 80i so it wouldn't cannibalize sales of higher-margin systems).

i would love to call this a regulation failure, but that would just provoke laughs

oh, possibly dumb question related to the protectli: they offer both AMI bios and coreboot. can illumos boot under coreboot?

probably a question for tsoome, but I think jclulow also knows

sommerfeld: Usually coreboot is still EFI at the end, right?

It's just a different, more open, implementation

jclulow: at least that's the theroy. Just wondering if anyone has tried it and got it to work. I'm cynically half-expecting there to be gotchas if you're booting something other than linux..

sommerfeld: I've a VP2420 and coreboot works well with SmartOS - much better than the AMI bios (because of some bug they might have fixed already with the serial console output)

I wouldn't even use "cynically" @sommerfeld.

that's basically the metric of cynicism, though.

drscream: thanks! will give it a shot then.

if anything, that's probably better given how much "value add" junk OEMs put in their UEFI implementations

I used to have PC Engines APU systems that they shipped with coreboot, I believe, and it worked OK after I fixed some stuff we were definitely doing wrong. illumos.org/issues/6859 I think

→ BUG 6859: missing MCFG table should lead to I/O PCIe config access (Closed)

like, don't add another incompatible layer of boot ordering beyond what UEFI recommends that you can't alter without unpublished specs