Hi. I am trying to get 9front iso booting in Bhyve on OmniOS. I have successfully did it 6 months ago, but now it fails to boot.

The console shows the iso starts, the 9front bootloader emits some info and then it attempts to start the kernel, but then the zone suddenly stops.

szilard - is there anything in the zone logs? (`zadm log <zone>` is the easiest way if you have zadm installed)

I see this: pastebin.com/raw/GB1uV9c2

That will do it, the assertion. We need to understand why it tripped and whether bhyve should handle whatever's going on better.

I have attempted to remove every optional stuff from the config, but zadm adds most of them back automatically. So I am unable to check which module / config entry triggers the bug.

It would be easier to look at the bhyve sources to see what's going on.

or dtrace, maybe

shouldn't this have left a core file?

I am looking at the pci_emul.c here: * Register (or unregister) the MMIO or I/O region associated with the BAR

* register 'idx' of an emulated pci device.

if you have a core, the stack will tell you which device failed

Woodstock: where should I look for the core file?

/var/cores?

I have many /var/cores/core.bhyve.13782

well, it's probably the newest one?

yep

deleted all, and bhyve created a new one, so it is definetely new

uploading right now

core.bhyve.20110 : 0x0.st/PHsX.20110

it is 10Mb

it's the framebuffer device

trying to register bar 1, which I presume is the actual framebuffer memory

afaik there is a bhyve config option to disable the fb. Or is this the efi fb? I don't really know too much about bhyve inner working.

If I disable vnc it doesn't crash, but it gets stuck

without vnc it hangs here: pastebin.com/raw/6ASdChgc

This is the consol output with BHYVE_DEBUG firmware

my scrollback buffer is too short to capture everything, sorry

in zadm log i see only this 2 lines:

root@omnios:/var/cores# zadm log 9front

wrmsr to register 0x401(0) on vcpu 0

wrmsr to register 0x401(0) on vcpu 1

I am trying it with a different OS iso to see if this is a 9front specific bug or not.

openbsd 78.iso boots just fine with enabled vnc.

Yes, 9front is obviously doing something we don't expect and we should see what we can do about improving things.

i don't recall the details, but working on vga bios stuff and gpu passthru a while ago i saw similar issues with bar remapping

Oh, this is a known issue : bugs.freebsd.org/bugzilla/show_bug.cgi?id=290098

Just checking to see if it's possible to disable the bootrom variables with our framework

Yes, szilard try `zadm set 9front uefivars=off`

andyf: this works. thanks!

I have successfully got it booted via vnc

Great :)

We should probably build bhyve with that extra detailed error message always enabled