jbk: okay, I misread my own code earlier, it makes sure not to run any HW emulation if the underlying part supports it, which I verified with a local setup and some dtrace. However, that code still should have been removed as part of the unwinding of OS-6778, but it's technically not hurting anything.

The reason you hit this gets back to the other issue I was talking about. Where a packet is coming down with HW flags which it shouldn't have

hmm.. i wonder how hard it'd be to figure out what's going on with that.. it's a separate issue to the thing I'm actually trying to work on...

I understand, I don't know if it's a configuration people ever have: multiple NICs on same subnet with differing hardware offloads

I can replicate such a setup locally, so I might see if I can replicate

jbk: `dtrace -qn 'e1000g_retrieve_context:return { @[arg1 ? "true" : "false"] = count(); }'`

with the assert remove you should see that return false sometimes

assuming you still do the SSH over e1000g0 and do something on the terminal to warrant SSHD sending an LSO segment

bahamat: fsnotify/fsnotify #12 is finally marked closed! Not sure if you still use telegraf, but if you are, perhaps improvements are finally unblocked?

Sweet!

Not at MNX, but I think Joyent still does.

But we do use prometheus, and it also uses fsnotify, so that'll be great for us.

hello

I tried to compile pkgconf and I got warnings saying that strdup() is not declared

according to this man page:

strdup() is POSIX

but in string.h, I see, like other functions, no POSIX guard , but __EXTENSIONS__

is it normal ?

[illumos-gate] 15077 ficl: make sure the loader emulator does set console -- Toomas Soome <tsoome⊙mc>

vtorri: looks like you are not setting up the correct C macro guards. Extensions is a special macro being defined with C99 flag I think but people here can elaborate which one you need.

[illumos-gate] 15063 vmm: cast to smaller integer type 'boolean_t' from 'void *' -- Toomas Soome <tsoome⊙mc>

[illumos-gate] 15027 The ls command should show SIDs instead of ephemeral IDs -- Gordon Ross <gwr⊙rc>

[illumos-gate] 15059 Update AMD microcode to 20220930 -- Andy Fiddaman <illumos⊙fn>

Are AMD CPUs still without illumos KVM/bhyve support

I know for certain AMD Zen and later works FINE on BHYVE using a modern build.

There may be more, but I'll defer to others for those details.

I run SmartOS on a Ryzen 5700G and BHYVE works just fine.

hasn't AMD support for bhyve always been there since bhyve support was added?

I think so.

Aaand soo, preferred hardware section on wiki for SmartOS AND Triton.. needs update to mention.. AMD works just fine for Bhyve..?

yes, even works great on my old crappy N36L

jperkin: I finally figured out how to bust up a JDK so that I could use the remote access card in my N36L haha

You have to turn of all the TLS safeties

jbk - yes, it's only bhyve PCI pass-through which is not available on AMD as far as I know

Who needs TLS safety anyway, right?

Gonna take a long-shot here, but does anyone have experience with AMD X470 chipsets spontaneously power-cycling after heavy SATA load (like "zpool scrub" on a 14TB mirror load)?

It's gotten LESS shitty with a BIOS upgrade in 2021, but it still happens 1/2 the time I think. And NO idea what's doing it. (I'm tempted to unplug it from power, re-plug, and try again, thinking it might be some insufficient BIOS-level counter.)

And now, with consistency, right before the powercycle:

ahci: [ID 517647 kern.warning] WARNING: ahci0: watchdog port 6 satapkt 0xfffffeb305ba9b18 timed out

jclulow: ah nice, I didn't bother getting that, is it one that needs a license to work too?

danmcd: let me see if we've upstreamed all of our relevant fixes yet

IIRC, the deadlock was fixed (though unlikely to hit that from a scrub.. taht was caused by out extensive use of 'artisinal SCSI packets' :P)

jbk you think it's an illumos problem? I can understand it might be because my dump device is on the very (big) drives being scrubbed.

there were some questionable bits with how timeout/reset handling was being done in the sata driver

i need to look to see the specifics

if they could potentially be involved

yeah.. i think there's a deadlock in sata_hba_event_notify.. we have a fix that farms some bits out to a taskq (it's not ideal, but better)

i'll try to get details in a bit once I can hopefully get this otehr stuff i'm working on finished

I may try and repartition my SSDs to have 128GB less unassigned space, and make those mirrored slices the "dump" pool whose only job is to be the dump device.

nvme?

(I do Weird Things with my SSDs...)

No, SATA.

Hand-me-downs that were new, but outlasted the kids' laptops they were installed in. :upside_down:

Pair of these:

0. c10t2d0 <Samsung-SSD 860 EVO 500GB-RVT02B6Q-465.76GB>

at least with the deadlock that I (finally) found.. one of the complications was that it locked up the sata framework enough that dumps wouldn't work to sata disks (which is arguably a bug, but haven't been able to dig into that)

Oof.

My symptoms were BLINK power gone.

oh that i haven't seen

I even put a video camera when I was able to reproduce it and NOTHING.

I need to restore my reddit because the only community with my mobo with any activity appears to be there.

The only complaint I have is that scrubbing (and also `zfs send -r` -ing) my 14TB mirrored pool (which is slightly less than half full right now) tickles this power-flicker thing.

Builds don't do it.

(I wonder if THAT's because those write to slog?)

Dan is there anything in the ipmi log by chance?

Lemme check again, but last time(s) there weren't, just "power on".

A lot of AM4 boards had VRMs that only barely meet the minimum requirements for Zen3

I'm running Zen2, FWIW.

@rmustacc kebe.com/~danmcd/webrevs/ipmi-log.png

Looks like power-off pair followed by power-on pair.

I've got "raw data" too, but that corresponds to reading my mobo manual, so that's on me.

Actually, no need. The ones above match reboot times to the T.

[illumos-gate] 15032 SMB: Can't create a file with a NULL DACL -- Matt Barden <mbarden⊙tc>

(The codes I'm seeing from the power-flicker match ones from me typing "reboot".)

That's unfortunate. the things that would come to mind that lead there that a colleague hit recently were due to a failing fan and thermal trip.

But I'd hope things like ipmi would do that.

"Thermal problems" *is* on my list of possibilities. My case is, unfortunately, not well suited for two disks like that to be so close to each other. I think it need a sufficiently long time to heat, though, and I've contemplated upping the fan count in there, or even replacing the case with a different one better suited to space between drives.

s/need/needs/

There case was due to a failed fan, so that seems less likely here.

[illumos-gate] 15050 SMB server mishandles some SIDs -- Matt Barden <mbarden⊙tc>

I have LWPs on the brain.

at least they're lightweight :P

jbk: eh

I specifically have the matter of "how do i make a thread-unsafe greenthreads system-thread-safe" on the brain

one big mutex

Erlang

that was for your arekinath

I feel like the BEAM answer to "how do I make it system-thread-safe" is "add more threads"

which is intuitively crazy but seems to somehow work

but also those people took too many puffs on the macro pipe

Rickard Green keeps your green threads safe

when all you have is a mailbox, everything is a message

jclulow, that's called a giant lock

arekinath, I mean...

Who would win in a fight, though, one giant lock or a giant pile of little locks

what if, hear me out, you have one giant lock to keep track of the little locks

the giant pile of little locks really know how to kill you dead(lock)

bah

TIL

programs with giant locks are considerably shorter than programs with fine-grained locking. here is a driver in 4.4BSD. and here is the equivalent driver in Linux

and some system are on crusade to eliminate it, e.g. WARNING: Device "psm" is Giant locked and may be deleted before FreeBSD 14.0.

it's the Bananarama of concurrency primitives

hm

I think the word you're after is supervise, danmcd

but you can spell it with a Z if you really want

the compiler will accept it

Sorry about that, I was dining with some philosophers earlier. (Now I have to leave... seriously.)

and dodge the produce? :)

¿_?

danmcd ran off with all the forks and now the rest of us must starve

does that make him a Byzantine fork thief or something

I hear he ran off to the Island of Paxos

this is some concurrent programming joke I don't understand, init?

Diabolical algorithms: a reader-writerS lock. In this algorithm, there may only be one reader at a time, but there may be any number of writers. I can't imagine this being useful for anything except efficiently, cleanly and quickly causing severe data corruption.