jclulow: part of me wonders if we should keep the actual definition of `struct ilstr` and `ilstr_flag` out of the "not-public public" `sys/ilstr.h`?

oh I guess it's not being built into a .so anyway is it

arekinath: I want to be able to drop one on the stack

so binary compat isn't a big deal

yeah ok

right

that's why it's like list.o not libcustr.so

do we have AMD EPYC "Rome" support?

neuroserve: Yes.

We make relatively extensive use of modern AMD CPUs (both consumer and enterprise parts) at Oxide. They work well.

just asking 'cause im sitting through an AMD roadshow right now - we're using "Milan" right now and it seems that "Genoa" or "Bergamo" might be next

bhyve pci passthru doesnt work on them though, or did that get fixed?

Not sure. We're not doing any of that at the moment

I'm trying to trace ipc calls but I don't see shmget in the list of probes with dtrace -l , this old post also checks for shmget dtrace-discuss.opensolaris.narkive.…gr5fjgm/dtrace-and-shared-memory-id to migrate this script to illumos which provider do I need ?

you can use just function names; probes are good to get specific information from inside of function (variable values etc)

shmget is definitely there: src.illumos.org/source/xref/illumos…/uts/common/os/shm.c?r=86ef0a63#829

tsoome thanks!

tsoome do you have any experience booting illumos under tiano core? I'm still debugging on oci how to boot smartos and clearly, it enters exitbootservices call but never returns

tiano core like edk2?

tsoome yes that's right

I have used qemu (cueerntly UTM)

currently*

I removed extibootservices to shutdown(1) to get back to efi and as expected I get back to the efi shell, but calling Exitbootservices never returns

exit boot services needs correct memory key

how do I check if I have the correct memory key? that's being return be efi_get_mmap I think

well, normally exit boot services should complain if the key is not ok

and yes, you get it with memory map

let me see...

what system is using this tiano core?

tsoome I'm trying to boot using ipxe on oracle cloud

ah, oracle cloud.

thats x86?

not aarch64?

ExitBootServices is using image handle and key, if key is ok, perhaps the image handle is corrupted?

I did work for ampere to fix edk2 used with their aarch64, that one was booting fbsd loader however.

tsoome oh, I have not check the img_handle I'll try to print it before the call

is there a magic header to check if it's corrupted?

anyhow, I'd suggest to try out illumos loader64.efi

um, not idea really:)

have had no reason to check whats inside there

I mean, if illumos loader seems to work, it means ipxe has something oing on. if both are bad, then there is a problem...

how positive you are that ExitBootServices does not return?

I'm asking because, if your console output in ipxe is done with SimpleTextOutput protocol, then you can not print anything out after ExitBootServices. loader does set has_boot_services = false and is using software rendering on framebuffer to get output (if needed - normally we get next output from kernel already)

if serial console is used with Serial IO, you have the same proble,

I compiled ipxe with debug messages, so there is debug message when bootservices fails and one after is called to print the return status

oh, so after calling exitbootservices I won't see any messages printed?

but if ExitBootServices did not fail and you attempt to use BS-> functions, your system most likely will hung

after ExitBootservices just multiboot2_bounce is called

that just trampoline to switch to 32-bit protected mode and jump to dboot?

I'll to check the multiboo2_bounce call implementation, at least I don't see it in the ipxe code

loader does relocate kernel and modules to final location, then does switch to 32-bit mode and jumps to kernel. relocation is needed because when BS active, we can read kernel only into allocated memory.

tsoome thanks, I'm just learning this as I tried to boot smartos on oci.

yw

tsoome I was wondering on what you told me about messages and now I just tested using serial console ( I was using vnc), and found out that was crashing, in vnc I never saw this.paste.ec/paste/CkHm3VpB#42-aIbq3o0Mtt7Ae4JQjir1FXFfKLASO886KjgVs2Ms seems is the image handler could be wrong?

ah, it is quite common that exceptions are dumping output on serial port and not on video

tsoome that exception means the image handler was incorrect?

from entry address you can see if this exception (invalid opcode) is happening in ipxe or in firmware

seems rather low for our kernel (unix entry address is 0xc00000)

but may it be this exception is result from your debug printout, no?

I removed the printfs, thinking that could the cause, I'll compile without debug

tsoome this is awesome, but sad because I spend like a week in vnc without seeing this paste.ec/paste/x54TT8Xo#XGMQ5zUMY1YB1d6BrCl+OMlKMgmF7+eVofKyt-SrOAn

do_bop_phys_alloc is the kernel right?

well I learned not to use vnc to troubleshoot things at least

jbk I could use mdb or is too early in the boot process ?

i think if you load kmdb at boot, you might be able to

i've not messed with that before though, so not sure if so or if you'd have to resort to printf() debugging

OS-7748

jlevon fixed this in our ipxe a while ago.

jbk, is to early for kmdb, is says loading kmdb then panics

(I was having problems booting a VMware guest over the network and we discovered this.)

kebe is pxe causing that we are not able to boot then?, I'll check if that fix is on the pxe version I used

ipxe not pxe

It was fixed in April 2022

oops

April 2020

(uggh, typing!

)

I *do* seem to recall John saying it could still happen even after this fix, given the "right" memory layout for a given EFI.

but seems the error message is different, it goes out of memory on the multiboot2 code, here is this is case is booting the kernel but panics on do_bop_phys_alloc, maybe I'm confused and is the issue is related

kebe the fix is there ab25dde3d656adfccecb04882b60849655e36be2 in ipxe

Yes. Correct.

kebe this has more information smartos.org/bugview/OS-8138 , that explains more things

Oh yes. You're attempting a SmartOS boot so this fix (which may need to be upstreamed?) is there, right?

Sorry that's all paging back into my head now.

kebe yes, commit ab25dde3d656adfccecb04882b60849655e36be2 in ipxe has all those tickets

And you're booting into SmartOS, not another distro, right?

kebe yes, I'm booting smartos using netboot.smartos.org/smartos.ipxe

Okay, so it's not, "Shit my distro needs this fix."

to do that I just copied the triton ipxe I compiled from github that according to gitlog has those fixes, then in the ipxe shell chain netboot.smartos.org/smartos.ipxe and choose the latest image from the menu

I'm assuming earlier images from the menu ALSO fail? (because if any of them succeed clearly there's a recent problem, even if it's something like "feature X bloats memory too much".)

I expect any image to fail, but it's a good test to run.

there are multiple ways it could still fail :/

Thank you for confirming my memory of our conversations from 2020.

(See 1059 US/Eastern comment...)

kebe let me try old images from the menu

kebe image 20221103T001803Z also failed, well this just seems to happen in oracle cloud if triton usually uses pxe to boot

Okay. I was kinda hoping the old ones would fail equally. Means we don't have to regression-hunt.

the dboot/early kernel memory allocations happen from memory after kernel+modules, I'm not whet the upper limit there is (memory hole/io space or something?)

okay so, illumos is not the system V I worked on at all

some of the userspace is likely close to the same

AIUI (and I didn't join Sun until the last days of 2.5.1 development), SunOS 5.0 branched from SysV once it was released under "Solaris 2".

(i.e. 1991)

aha, i worked for IBM in 1993, DEC in 1995, Tandem in 1996 and a couple of others in between

Tandem swore we were using SVR4 code

TLB2 i think they called it

but it was really close to unixware code i worked on in other projects

but kebe that would explain it

the yacc code i still haven't fixed looks like it's barely fixed from 1970s yacc

the locking code is much modernized

like much less locking

KS_HOLD0LOCKS() was a common svr4 kernel thing

(kernel space, i hold no locks)

anyway the code looks much nicer

and att/novell code never had documentation

you had to ask random people if they understood what you were working on

this is actually much nicer than when I was at Intel, *someone* usually understands what I am working on

kebe I made a change to test and now seems that is booting

neirac, yay for booting!

any progress on getting risc-v hardware to this project?

dangergrrl :)

kebe I'll try kmdb now as pit is broken in oci

neirac, I wasn't joking, risc-v would be perfect for illumos but i can't afford the stuff

the linux laptop is like $2k

and there was the announcement of 128 core server risc-v chips

i can do assembly!

someone needs to donate hardware even if it's in sweden

btw, this software is barely less complicated than old AIX

anyway, dropping sparc, maybe a good idea, dropping portability is not

I don't think anyone has suggested we drop portability.

However, just because you ask in the ether for other people to do work doesn't mean it will happen.

rmustacc, sparc was dropped

Correct, we did, but not because of portability.

i'm not asking for anyone to port to anything

i just don't have the hardware for another suitable target

ARM is not

Use a VM.

You can use qemu just fine to get going if it's something you actually care about.

Get to the point that the overhead of binary translation is actually a problem.

I actually have a Zedboard

but i don't think it will run rv64

I just don't have enough gates

on my fpga

i don't want this to be flatlined just because x64 dominates the market right now

rv64 is the future of OSS

so, port to qemu? just like i am a hardware dev... that's an option

I'm not trying to flatline it. Just trying to point out that you asking folks to do itdoesn't make it happen.

If you want to, then great.

Using emulation is the classic approach fwiw. Even was true of unix ports a long time ago.

this is unix

i have classic differences with sun

but this is unix

not sure how that's relevant?

i will run away and run linux if you ask

emulation is the classic approach and you win

All I was trying to do was point out an approach to make progress on something you wanted to do.

and i wasn't asking you to port to rv64

This isn't a contest about winning, losing, or telling you what to do.

you asked if the project had done it and with a tone of well why not.

just trying to figure out i can iwth the limited resources i have

I'd recommend getting to know the existing system, fix some bugs, and actually use it.

ok, i suggested you might want to

i use it

i have posted bugs

i have not fixed the bug in yacc i was assigned

ok, accepted

I use it

it doesn't perform over linux for me

but linux has 10G people

i am patient

If it doesn't perform the way you want, there's a great place to dig in and improve things.

the non canonical stuff bothers me but linux is more noncanonical

i like the init replacement better than systemd

Our goal isn't to be a canonical unix.

ok is related to OS-8138 , I modified the 1gb limit and smartos boots but hangs on the pit issue, but I cannot use kmdb because If I do it it does not boot or takes so long that

danke

but unix was something i helped make when i was young so i will always be attached to it

forgive me

i have not read this projects goals

so perhaps i should

There's nothing to forgive, don't worry about it.

is there a link, please or should i just google it?

if i don't know the goals, i don't know if i align with them, i just saw really cool software that i helped create and i volunteered

this is mostly a really good server OS except it fails with modern windows and OSX

the guy who was doing that work was caught in a war... that sucks

can i have his bug/defect?

I would start at illumos.org/docs/about and you can read from there.

What issue number are you referring to?

neirac: OK, so we're back to the point where getting something in as an exception for whatever is going on for the pit would help?

rmustacc now booting using pxe works, if I modify the 1GB limit for allocation for ipxe to 512mb I just tested with an arbitrary number smartos.org/bugview/OS-8138, the thing is that now illumos boots but is stuck as pit is broken , but I still cannot go into kmdb because choosing that option from the pxe menu it does not loads the kernel is frozen, I'll try using a smartos.ipxe image that has pit_is_broken hardcoded to 1

rmustacc just to rule out that is stuck by other reason but at least there is progress

neirac: can I ask why you're using PXE here and not just pushing a regular cloud disk image of some kind?

jclulow sure, I'm on the free tier and cannot use custom images

Ahh

What a ridiculous restriction

jclulow yes, things would be easier, I think the idea is to force the move to a paid account instead of just using the free one

Wouldn't expect anything less from ORCL...

Probably I would boot an existing Linux image and just dd my disk image over the contents and reboot then haha

And honestly a free-tier from ANYONE is gonna do things like that.

kebe: Only at places that don't respect open source developers trying to do platform enablement work

jclulow we tried dd the iso to the disk, but the smartos installer fails when doing so

it complains about the partitions

neirac: I wouldn't dd the ISO, I would prepare an installed system image somewhere else

jclulow beat me to the punch.

And dd that onto the disk

Esp. SmartOS with a bootable `zones`, you need to do that.

After finding some blog post about how to pivot root a Linux system so that you can unmount / and overwrite it

I just saw one of openbsd, but they imply dd to the disk as susual, is just smartos complains on partitions when the installer is going to create the pool

I take it this VM boots EFI not BIOS?

kebe yes efi

So the disk-bootable zones pool needs to have been created with `zpool create -B`, which the modern SmartOS installer will do.

Josh's advice applies, just make sure you grab the whole disk.

For this example 32GB bootable zones pool disk:

* Partition Tag Flags Sector Count Sector Mount Directory

0 12 00 256 524288 524543

1 4 00 524544 66567903 67092446

8 11 00 67092447 16384 67108830

partition 0 from prtvtoc(8) is the EFI System Partition (ESP).

so I could just dd smartos, the. go to a shell and type zpool create -B

You also need to make sure OUR loader is in the ESP, though. This is all stuff `piadm bootable -e` does.

If you make a SmartOS disk image that works in, say, QEMU

booting with UEFI on the serial port etc

And then you take that whole disk image and dd it to /dev/sda or whatever the whole disk device is in your Linux

I suspect it will work about as well as anything else

I would probably either match the size of the target disk with the image that is being created, or zero the parts of the disk that your image does not cover

(there are backup UEFI labels at the end of the device that you'll want to either overwrite or clear out that could confuse things)

jclulow thanks I'll try that, I'll try omnios also I have not tried that

um, have I missed something about RTI of 15227 and 15255 or are people just busy otherwise?:)

hi tsoome and thank you for your contributions to illumos this year

sorry, i know it's off topic

ou, thanks:)

it's my favorite foss project

.oO same procedure as every year..

i want to tank jclulow and danmcd too

fortunately foss is something you can spend whole life on;)

tsoome, i hope i am allowed to thank people for the holidays

[illumos-gate] 15222 pcieadm pcie.slotctl.pwrflt description is misleading -- Robert Mustacchi <rm⊙fo>

tsoome, unless you are larry ellison

I'm sure you are:)

ok, lets see why those acl tests are failing...

sorry but he seems like the grinch

you won't even discuss him :)

na, if you cant change it, no need to worry about it;)

well, i didn't work there and i hate larry dor what he did to other teams

tsoome: I'll go look.

thanks!:)

lol

my friend suggests we destroy larry by having him buy twitter at 20BN

sorry but it was funny

then there would be poll about if he should step back....

tsoome: On 15227, wasn't sure if kebe was planning to follow up or not. 15255 replied to.

yep, thanks, I just have to verify if I actually have that patch built+installed already:)

jinni illumos 15227

fenix illumos 15227

BUG 15227: libsmb: variable is used uninitialized (Pending RTI)

↳ illumos.org/issues/15227 | code.illumos.org/c/illumos-gate/+/2539

Sorry about that tsoome...

no worries!

You have new mail.

i need gugs

my buyout of oracle failed

thanks!:)

oops

sorry

I still have my side of work to be done with few others:)

i will apologize and keep humor limited

[illumos-gate] 15227 libsmb: variable is used uninitialized -- Toomas Soome <tsoome⊙mc>

[illumos-gate] 15255 loader: replace nbsp in forth scripts by space -- Toomas Soome <tsoome⊙mc>

code.illumos.org/c/illumos-gate/+/2518 is still needing reviews, then all libdladm cleanups are reviewed:)

→ CODE REVIEW 2518: 15196 libdladm: comparison of integers of different signs (NEW) | illumos.org/issues/15196

hey there! are there any plans to sync with OpenZFS?

it happens as things are wanted/desired/time permits

faster for bug fixes etc, usually.