a question a bit into the weeds, but does anyone know offhand w/ linker sets, what is the specific mechanism that makes the __start_set_<setname> and __stop_set_<setname> symbols appear at the start/end of the section it creates?

or i guess that guarantees that that happens

jbk linker does. But, depending on build system, linker maps also have a role there (if we are telling which symbols are to be preserved, we most likely want to make sure our sets are preserved).

.oO paste.ec/paste/GRhvtmzj#08CkEPgJEtMzprsXMR2rHB7+yTIG37wPfPG0hNFHz5r

Hello. I am trying to buy some SPARC hardware for testing OSes. Thought the kernel devs might offer some insights.

I'm not sure if I should get two T4-1 boxes (for 260 €) with 32 GiB RAM each, or just one M4000 with 64 GiB RAM (for 300€).

M4000 seems to offer 4x less threads/core, but total core count = 16 is the same as in total of two T4-1 machines.

If you're thinking illumos, it works great on the T4-1 (both Tribblix and OI are fine)

The M-series have been problematic

ptribble: Thanks. What are the M-series problems, if you please?

The problem is that it doesn't boot, it gets started and then just hangs, from the couple of people I know who have tried it

Thank you, noted.

if a nic loans up an mblk_t (i.e. uses desballoc()), at some point, it likely needs to wait until all such loaned up mblk_ts have been freed and an expectation that consumers upstack will do this at some point

but should barrier be the XXX_m_stop() routine, or driver detach?

jbk: I generally do detach.

Hey all, anybody who knows about nvme drives, I have a question.

When I replace an nvme device, usually cfgadm will show it connected/uncofigured, then I can cfgadm -c configure <ap_id>, and from there I can zpool whatever, which feels pretty normal.

But I've seen a newly inserted device show up in cfgadm as disconnected/uncofigured.

How do I change it from disconnected/unconfigured to connected/unconfigured so I can configure it?

Or does anyone have any pointers to where I can look to possibly make some progress?

I don't have any pointers. But isn't the model completely different from SAS/SATA?

E.g. it has DMA, it's reliant on PCI express hotplug working correctly

Yeah, we've got hotplug enabled.

-c configure would be the option, but it is another question, if it is actually implemented:)

Yes.

err -c connect

No, you want to go straight to configure.

bahamat: What's the hardware platform this is on?

We put back work that should auto-online PCIe devices like that.

rmustacc: It's smartos. I think the customer has intel processors, but I'm not 1005 sure.

Certainly the NVMe hotplug auto-online is working on our systems and was working on the i86pc systems I tested on.

Does it matter if it's intel or amd?

Somewhat, but only in so far as it changes the bridge settings.

Yeah, my test system it's working, but the customer is reporting the device is disconnected.

What SmartOS release are they on?

Good question.

Do you know any relevant bugs that I can refer to to make sure they meet the minimum?

Pulling them up.

→ BUG 15942: PCIe hotplug insertion does not configure the slot (Closed) | code.illumos.org/c/illumos-gate/+/3057

So first, I'd see if they have that.

rmustacc: Excellent, that helps. I'll make sure they've got at least that, and then see where we're at.

That may just be it.

Once you have that,a the next thing I would ask for is the bridge config space and specifically its slot capabilities.

If it's not that.

But what you're describing is the behavior I'd expect without that change present.

And you'd need to explicitly issue a cfgadm -c configure.

How do I figure out the bridge config space and slot capabilities?

/usr/lib/pci/pcieadm show-cfgspace will get it. If in doubt, just have them do a save-cfgspace -a and zip that up.

Awesome. I'll have them do that.

[illumos-gate] 5798 fexecve() needed per POSIX 2008 -- Andy Fiddaman <illumos⊙fn>

anyone know why when forking illumos-omnios in github, if I want to fork joyent-illumos I cannot ? it says that I already have a fork.

I imagine it's because they are both already forks of the same thing - illumos-gate

You can just have multiple branches in the one fork - one for SmartOS and one for OmniOS for example

andyf_ thanks a lot, I'll do that

tsoome: have you seen 'zf_read: fill error' before in loader?

when loading the boot archive (at least the error happens between the message for that and boot_archive.hash)

zf_read is failure of decompressing gz file -- boot_archive I guess

our boot archive is gzipped and will be uncompressed in memory

ahh was shitty server firmware that is buggy w/ firefox

rmustacc: (since you probably know) the 'polled' callbacks for keyboard drivers are basically for kmdb and need to work w/o interrupts, correct?

I would expect so, yes.

Or some other system state.

Certainly for the USB part that is true, but there are specific hcdi entry points for that. I don't remember how that looks at the individual driver.

Where are you?

[illumos-gate] 16118 Fix problems with libnwam communication between 32-bit and 64-bit applications -- Carsten Grzemba <cgrzemba⊙oo>

I have a RT5360 Wireless 802.11n 1T/1R, that works with the ral driver when I do the following sudo add_drv -i 'pci1814,5360.1186.3c05.0' ral , at least I dladm show-phys shows it now. here is my prtconf -vp info termbin.com/2irc

just trying to get some comments correct for the hyper-v 'enlightened' keyboard driver

(at least initially the polled keyboard callbacks are stubs)

what do I need to change to make it attach the card to the driver, without using add_drv -i ?

oh it seems the driver needs updating, just running dladm scan-wifi fails, dladm: cannot scan link 'ral0': operation failed

our 80211 stack probably needs updating as well..

i thought a neat project might be to port it (or just start w/ the freebsd one -- i think that's where ours came from way back, just not updated in forever) and convert it to rust, but ENOTIME

since it's not as critical as other stuff, but the kernel bits have to do lot more with what is basically untrusted data coming off the wire

compared to say an ethernet driver which (at most) _might_ need to read the ethernet header off the wire (and most don't even need to do that -- just pass upstack whatever it gets off the wire)

jbk thanks!, I thought it was just a matter of updating ral to make it work at least with the card I have

i'm not even sure offhand if our stack supports g or n

rmustacc andyf_ since you guys work with TB of RAM, I was wondering if you've ever tried provisioning a VM with 1TB+ of RAM on illumos? I can have a 1.5TB RAM VM on FreeBSD, but I've been wondering about moving the host to illumos. the VM is ubuntu, btw.