antranigv: COntributions needed github.com/richlowe/arm64-gate

I am guessing lofiadm does not work in a zone?

it should.

hmmm then I need to debug why it failed

toasterson as much as I admire illumos and look up to it (and port amazing features back to FreeBSD :P), I don't have enough knowledge to contirbute. unless there's a mentorship program, then I'll look into moving my whole company to illumos (including our custom OS :D)

antranigv: I do consulting for that kind of large moves.

toasterson can I DM? :)

Sure

tsoome: Well then I have found a bug. Or at least a misconfiguration. lofiadm: /dev/dsk/c4t1d0p0 was not created: No such file or directory

It shows the disk in the output... (full message at <libera.ems.host/_matrix/media/v3/do…cec91e959f578e40cf4020e1611dc409dd0>)

but there is no disk link

you probably need to add this device name

/dev/lofi/* are there, but disk names are not.

you mean to the zone perms?

yes, in zone config

is there a need for lofiadm to use /dev/dsk instead of /dev/lofi?

only if you need disk emulation for partitioning etc

ah, ok so yes I need to add that then.

lofi = loop file?

yes

right

hmmm, something seems to take to long for lofiadm. it creates the /dev/dsk/c4t1d0p0 now but lofiadm tries to stat64 for it and gets ENOENT

so lofiadm fails

oh, because I am missing rdsk....

in zonecfg perms

ah:D

I was already suspecting some new and fancy bug:D

about bhyve and booting solaris 11.4 CBE (the free developer version). it unfortunatley hands while loading modules (reaches "dld0" then freeze for ever). does anyone have an idea how to debug this?

it would be a small help if debuggung the module load could be more verbose, e.g. getting printed to console "start loadin module asy, enteriny asy, ..."

is there a kernel setting that makes module loading a bit more chatty?

tomww: boot -kvd moddebug=0x80000001

and take a look what stopped

moddebug=0x80000001 - after loads to kmd

Ahh, thanks. It's been several years back when I last used the extra setting. I'll try and tell if that shed more light at the problem.

okay. maybe on a track. it fails a lot earlier, load 'ska/ska_kpti' id 5 loaded @ 0xfffffffffdc00000/0xfffffffffde00000 size 86224/8288

and I'll play around a bit. igork thank you!

why create zonecfg_get_dflt_sched_class() taking size argument of int when all call instances are providing result from size_t.... grrrr.

because it was 32bit and nobody noticed

believe me, I have met a lot.

that sounds like the qmail bug... :P

if you're fighting this too tsoome, check out the arm gate

I do believe you:)

it's possible I "fixed" a lot of things you'll meet by the compiler.

(I assume this is compiler stuff, not 64bit-izing stuff?)

'cos if youre doing that there's way more pitfalls, and shouldn't upstream what's in arm (as the readme tells you not to) :)

tsoome: I am not that consistent :)

well, this specific case ends up in warning about this comparre: if (strlcpy(class, sched_str, clsize) >= clsize)

where clsize is int...

igork: tried more switches from lars.timmann.de/wiki/index.php?title=Solaris_kernel_debugging

whee, zonecfg_get_dflt_sched_class is private interface, so we can fix it:)

You can probably fix it regardless, because the actual problems are going to happen either way

the problem would be types that potentially get smaller

(don't quote me on that though)

:D

tomww: once you're multi-threaded (which happens really early), the 1 bit of moddebug gets less useful

there's a setting -- somewhere -- that disables multi-threaded autoconf

but I don't remember where