it's also worth noting that the IP interface has an MTU that starts from the underlying datalink, but is truly separate. That said, the question of whether it binds or copies is dependent on several things, such as the number of mblks, etc.

The choice to bind / copy is done well before that point though. The ring tx path calls mlxcx_buf_bind_or_copy() which iterates over the mblk chain.

I would look to see some of the stats and DTrace probes that are there to help figure out what's going on.

The only copying that takes place in the aforementioned function is for the headers when they're inlined into the descriptor. Otherwise the descriptor chain is all set up in advance of that.

yuripv: Sorry for that delay. Running a safety nightly now before I push both.

gwr: while looking at something unrelated, I came across this which might be relevant to your question: src.illumos.org/source/xref/illumos…/stand/lib/tcp/tcp.c?r=b531f6d1#632

(it does make me wonder what the impact of that on things like LRO might be)

danmcd: no worries, thank you

[illumos-gate] 13313 tools build uses cw from /opt/onbld even if -t flag is specified -- Yuri Pankov <ypankov⊙tc>

[illumos-gate] 15012 nightly should fail if tools build failed -- Yuri Pankov <ypankov⊙tc>

Good timing...

:)

How often does that allocb fail, jbk?

Oof, also it uses "0" as PRI, which is unreadable AF.

err

too early still

ENOCAFFEINE perhaps. I still stand by using 0 as priority is Bad.

yeah, but that's a separate tcp stack :) which i realized later

which uses a lot of the same function names as the 'real' tcp stack

and at the very real danger of ENOCAFFEINE twice...

i don't suppose anyone still has a setup where bhyve<->non-VM zone still hangs regularly... I was never able to reproduce it locally

but while hunting down something unrelated (mostly just curious)... I found something that'd be nice to at least eliminate

(it still annoys me i was never able to chase that down)

i'm having lots of pkg issues on hipster, e.g. cannot install git, cannot update. is that expected?

what sort of errors are you getting?

can't gerrit auto-close the reviews once it's integrated? "abandon" feels wrong somehow

install git gives "This version is excluded by installed incorporation consolidation/userland/userland-incorporation⊙0"

update gives "No version for 'incorporate' dependency on system/library/processor⊙0 can be found"

trying to run "pkg update -v pkg://openindiana.org/consolidation/userland/userland-incorporation⊙0:20221009T143209Z" encounters lots of No version matching 'require' dependency errors

looks like "noincorp" was forgotten when obsoletting system/library/processor?

Has anyone created a prometheus node_exporter that supports illumos?

well, node_exporter runs on illumos and we have a package for it..

or was there something specific you needed?

nah, just curious that it ran

yeh it seems to work fine and at least gives basic OS metrics

yuripv: seems that way, I will create a PR and see what happens. Any advice for the "pkg install git" issue? "This version is excluded by installed incorporation consolidation/userland/userland-incorporation⊙0"

Am I supposed to do a change-facet or just remove the incorporation? (im new to all this)

toppk: After installing/upgraqding with IPS pkg manager, you most probably need to reboot into the boot environment.

toppk: (new boot environment)

kernspin: I should have mentioned, the first update updated 27 packages, and seemed to work fine, but new BE cannot do another update. neither the original nor the new BE could install git

toppk: Ok!

tsoome: I'm seeing a slew of these clang tickets with 'unix' in the title when they really apply to a different module

which makes it seem like they apply to the core 'unix' portion of the kernel, rather than separately loaded modules

pmooney oh, vmm is built separate modure, my bad, I'll fix it.

thanks

pmooney I was running make in i86pc ;)

its built under i86pc because it reaches into some machine-arch stuff

but in the end it emits a separate module

yep, there are some...

probably applies to 15066 as well (although being grouped under pcie)

yep

errmm, what's this one mean: pastebin.com/MefHdS5F

system/library/processor is obsolete, and it seems confused

perhaps something still depends on it somewhere?

or could do, in whatever pkg is trying to achieve for you

just doing "update"

I did update earlier, and it seemed...suspiciously small

so after the reboot I tried again

image-update is just an alias to update, no? Nothing different?

yeah, nothing different.

pkg search for things that need that package might help? if that's what's making it not just remove it

how do I find dependent packages?

pkg(8) has examples of searching based on dependencis, I never remember it

:depend:... something

if you search the manual for :depend: it'll show up

or depend:?

returns empty:

then I'm out of immediate ideas :(

richlowe: so, doing -vvv does give more info: pastebin.com/70BpwQ6z

I have no idea what could cause you to have SUNW* packages still, so that looks weird.

is it safe for me to remove system/library/processor?

it seems the apache package on there is needed for system-repository

I don't think it will work, or help

the SUNW* one is?

no, web/server/apache24, which comes up under pkg list

this seems like something you'd want to talk to #oi-dev about

yeah, it's the stuff about SUNWapch22 and whatnot I don't get

pretty quiet in there :-/

only other place I know of is the mailing list unfortunately

hmm, well, mailed the ML

hmm, seemingly not alone

[illumos-gate] 15040 remove support for unused sun studio inline asm -- Richard Lowe <richlowe⊙rn>