has anyone come across this device before? pci1d6a,1 (pciex1d6a,4c0) [Aquantia Corp. AQtion AQC113 NBase-T/IEEE 802.3an Ethernet Controller [Antigua 10G]]

seems they were bought by marvell but there are freebsd driver sources available github.com/Aquantia/aqtion-freebsd

yes, even in-tree in 16-CURRENT

a friend of mine had aquantia in his home pc-router, with freebsd.

ah nice! Here there are two of those in a "ugreen NASsync dxp8800+". For now I've just plugged in a supported pcie card, otherwise the device seems to run well with illumos :D

away from my illmos box atm, does /usr/bin/env support -S?

omnios manpage doesn't mention it. Got a test?

anything should do /usr/bin/env -S true

-S appears to be a GNUism

/usr/gnu/bin/env might work

Originally from FreeBSB I believe, but adopted by GNU

It does not work with /usr/bin/env but does work with /usr/gnu/bin/env on OmniOS 151054

OK thanks

by any chance has anyone tried to use lightningcss via npm on illumos?

it seems to be some sort of unholy chimera of node and rust

jperkin: I think your mpidr changes aren't needed now, can you close the PR if you agree?

andyf: unrelated in fact but not spirit to your make/spawn work. I have wanted, since sommerfeld showed me, to introduce Makefile.subdirs and use make -Cw in it so that make can 'exec' without a shell, and avoid `cd $(@); pwd; $(MAKE) $(TARGET)` (which has to be shelled out, because ';' is one of the magic characters)

it is almost certainly not _actually_ heavy though

The conversion of `dmake` to use posix_spawn shaved 4% off a full gate build (11% off system time) on my AMD Rome. That's more than I expected as it's not that spawn heavy, I assumed.

oh wow

Switching to posix-spawn-in-kernel on top of that didn't make much difference for that workload.

After this week's smartos-live release goes through the testing grind I'd like to run my "build all four combos" tests where each one has any changes in this space you have. Since SmartOS uses in-tree dmake, maybe the dmake-only changes don't even need a machine reboot?

andyf: how does gcc spawn children? there's a fair number of processes below the compiler frontend too

if you're actually hunting seconds in builds

danmcd - they do not, you just need to replace /ust/bin/make (dmake is a symlink to it)

richlowe I'm not really hunting seconds, it's a side quest from the posix_spawn stuff.

but gcc is worth a look.

danmcd - code.illumos.org/c/illumos-gate/+/4805

→ CODE REVIEW 4805: 18161 dmake(1) could use posix_spawn() to run commands (NEW) | illumos.org/issues/18161

Hmmm....

It's always fun matching the style of some of the older code :)

I would not be against, in the worst cases (like I tried with libm, but got tangled in reviews...) an automated stylistic pass as a separate commit

if you wanted to clang-format, or uncrustify, into something more illumos-y

and while me not being against it isn't so important, core weren't against it when I tried...

richlowe: done, thanks

andyf: I'd love posix_spawn everywhere, made a clear difference when I patched mksh and bmake on macOS for pkgsrc builds

from a _quick_ glance, it appears gcc uses vfork() + execv[p]()

uugh..

why does it still default to 4.4.4

jperkin - the main win for dmake was the switch from fork() to vfork(), since it's still using the libc posix_spawn.

ok.. so it did

jbk - probably not too bad; at least it's vfork!

but if anything's very multi-threaded, the kernel stuff will help when it's ready (and a switch to posix_spawn to make use of it).

it looks like it might use it in at least 15

though there's some other voodoo going on