I now got to get my wifi working. I got to get my wifi arn0 up in ifconfig. I have plumbed it. and I have tried running ifconfig arn0 up. But it still shows down in dladm. Tommorow's problem I guess.

[illumos-gate] 15504 ZFS/libshare is too verbose when services are not present -- Andy Fiddaman <illumos⊙fn>

[illumos-gate] 15500 tgtmap(9) tgt_privp doesn't work as advertised -- Hans Rosenfeld <rosenfeld⊙gho>

is it correct to not see vioblk interrupts ? termbin.com/wi1m.

do you have any vioblk devices?

I think you should see 'vioblk_int_handler' if any vioblk devices have attached

termbin.com/nml4 that's prtconf -d I thought having vioscsi meant having vioblk

no, they are two different types of devices

vioblk uses its own (simplified) set of commands (read, write, etc) for a storage device

while vioscsi uses the SCSI command set (far, far more complex)

jbk oh ok, thank you!

jbk is it possible to use fixed interrupts for vioscsi ? termbin.com/16u1

neirac: MSI-X is really what you want. Why do you need fixed interrupts?

jclulow I was reading this ticket illumos.org/issues/12015 , trying to find out why trying smartos in oracle cloud fails when writing to the disk(my assumption) as it not possible to create a zpool

→ BUG 12015: vioif with MSI-X not working on Google Compute Engine (Closed)

I would not assume it is related

GCE has a totally custom hypervisor thing and it has specific quirks

jclulow ohh, ok. the issue is totally different then. I'll try to troubleshoot more time the disk

Yeah, I had to do a fair amount of interactive debugging to figure that out

Which at least was possible because GCE provides a decent serial console!

Hopefully Oracle cloud does too

jclulow oh yes, I'm in a serial console, I boot through ipxe

Great

I assume this should work on any disk prtvtoc: /dev/rdsk/c1t0d1s0, but I'm getting this Unable to read Disk geometry errno = 0x5

It might not work if it hasn't ever been labelled? I can't remember

Try to give it an EFI label with format -e

neirac: I'm pretty sure that's what you get if there is no label.

fastest way to create label is zpool create ..; zpool destroy.

tsoome I tried zpool labelclear but failed : failed to open /dev/dsk/c1t0d1s0: I/O error

not labelclear, create.

we create disk label with wholedisk setup;)

tsoome zpool create c1t0d1 cannot label 'c1t0d1': try using fdisk(8) and then provide a specific slice

wot. hm what is device tree for it?

device path*

ls -l /dev/dsk/c1t0d1s0 would show you link to /devices tree

or format will show it in disk list

tsoome lrwxrwxrwx 1 root root 57 Mar 21 13:52 /dev/dsk/c1t0d1s0 -> ../../devices/pci@0,0/pci108e,8@4/iport@iport0/disk@0,1:a

anyhow, for EFI label, you would need to create 100% EFI partition in fdisk (either run directly or format -> fdisk)

iport? sas?

this is vioscsi, I presume.

tsoome yes, is vioscsi

I'm on thin ice there, but vioscsi does not provide cmlb_tg_ops_t -- I'm not sure if it is supposed to or if there is something else stacked on vioscsi

cmlb_tg_ops_t will provide callbacks for disk label (cmlb) from block device

looks like I've used 'fdisk -B' successfully in the past to add/replace labels on drives before adding them to a zpool.

err

fdisk not required?

format -e will create an EFI label if you ask it

I honestly don't remember why that was a necessary thing, just in my notes that I did it.

tsoome: sd provides cmlb

it's SCSI, so vioscsi is just the HBA

sd is the disk target driver that would sit on top

Because it's all not working right, I would use format -e on the disk and try to create the label

and then observe why that isn't working

Some sytem call like open(), or ioctl(), or even some read()/write() will be failing

*system

format has its own quirks.

It's baroque but it's fine

It should work here, for this, at least, and if it doesn't, there is a bug

we have added support for some devices in format in past

ok, I'll try format now

vioscsi should just be sd disks, though, not a new type

format failed termbin.com/s4zf

oof

Current Disk is unformatted.

that is ... not ideal haha

You should figure out why we think that

this part actually seems to work at least: c1t0d1 <ORACLE-BlockVolume-1.0-46.58GB>

so we're getting some of the SCSI pages that identify the disk vendor/model and capacity

FWIW, you should not have to use the "format" command there

Especially not for a virtual disk

jclulow at least I know what to check now

Yup!

That "is unformatted" message will come from something in format, which will have been a decision based on some ioctl or whatevr

and ... you'll be able to trace that back to something sd got from the disk basically

at least it looks merely like a SCSI thing, and not some interrupts bullshit

interrupts are the worst hhaa

*haha

jclulow thanks!!

You're welcome!

[illumos-gate] 15506 zfs and zpool tests can fail because of core size limit -- Toomas Soome <tsoome⊙mc>

I just compiled format with developer menu but I don't know how to translate the info to be good or bad termbin.com/asyy