i40e, ice, and others of its ilk (my recollection is that bnxt started this trend) is that it's the HW-RAID of networking. "Can't you just let crossbow do all of this?" is networking's equivalent of, "Can't you just let ZFS do all of this?"

I appreciate anytime I set the MTU to 9000 and it works.... This stuff is complicated lol

I started to read about "virtual station interface" rofl

Downloading the Datasheet and trying to pretend to understand :)

which?

the x710 statics.teams.cdn.office.net/evergr…sets/safelinks/2/atp-safelinks.html

sigh.... thats what I get for copying from TEAMS

errrrrrr intel.com/content/www/us/en/support…es/000093196/ethernet-products.html

you know, there are easier cures for insomnia :)

bahahahahah

Its helped with exercise...., I thought I'd rather do anything other than run ....

i mean, at the core if like practically every NIC, there's one or more rings, which is just an array (just when you write the last entry, you wrap around and start writing at the first one, hence the term)

and basically every element has a field for the physically address of the data to send and a length, and usuaully some other flags depending on features

often one to indicate which entry is the last one of the packet (so a packet can be split across multiple entries)

then you write the index of the first packet somewhere to send

and on the receive side, you just write the physical address and size of a bunch of memory you preallocate, then when a packet comes in, there's an interrupt and you can read a register that says which entry is the first one (and again usually there's a flag to say the last one)

or some variation

it's just all the _other_ stuff you have to do to get to that point

(which for those cards is quite a bit)

plus they have a built in network switch, so those different rings can basically at like virtual NICs on their own and do switching

(but then you have to worry about well which ones are in the same broadcast domain.. which mac addresses are allowed where, etc)

and for the intel nics, you also have to deal with their horrible LSO implementation :)

(horrible in the sense the amount of work you have to put in to use it almost gets to the point of not being worth it just because you've done like 70% of the work already)

And I'm guessing the affinity for the chelsio cards comes from them not being as much work?

unfortuantely, i've not had a chance to use them

but it does seem like they're working with those that are using them

so I can't say

but i'd hope their LSO support isn't nearly as annoying as Intel's

though I understand they can do quite a few other offloads if you want...

(I kinda wish we'd look at them... just like someone wants dell or hp, and basically they both go 'your choices are broadcom or intel'

[illumos-gate] 18220 mdb: 'end' may be used uninitialized -- Toomas Soome <tsoome⊙mc>

any crontab experts here? I'm trying to understand this part of the crontab(1) manpage:

The sixth field of a line in a crontab file is a string that is

executed by the shell at the specified times. A percent character in

this field (unless escaped by \) is translated to a NEWLINE character.

I'm hoping someone, somewhere, has an example. My goal is to have a job fire at 01:42 on the first Saturday of the month.

I have first Sunday of the month jobs like:

0 1 1-7 * * [ `date +\%a` = Sun ] && zpool scrub rpool

I'm testing with: 03 10 * * 3 [ 'date +\%d' -le 8 ]

but I'm getting email saying [: missing or similar. When I add : to the [ I get an error about [:: not found.

oh bother, never mind. Google was leading me way, way astray.

forget I said anything. I hid my head in shame.

Yours seems to be missing the space after the [ and using ' instead of ` (backticks)

Google was implying that the sixth field was part of the cron smarts, not just the script-to-be-executed.

so I was barking up the wrong forest.

(as in, the ai summary was totally messed up and I was messed up for trying to understand it)

andyf, your job is *exactly* what I'm trying to do, just different day. :)

i should get back to working on svc.periodicd at some point... just between the xml parsing, xml manifest additions, and libscf pieces, it's sooooooo tedious i can only manage small bits at a time :)

It's a good thing I have a meeting now. I can't do anything else really stupid for at least an hour.

[illumos-gate] 18003 poll(4D) needs to better-check numfds (CVE-2026-35384) -- Andy Fiddaman <illumos⊙fn>

[illumos-gate] 18020 Double copyin of dld_ioc consumers -- Dan McDonald <danmcd⊙ei>

[illumos-gate] 18117 SCTP needs to better-check INIT ACK chunk parameters -- Dan McDonald <danmcd⊙ei>

Start your distro builds folks!

heh..

was doing that as you typed :)

granted, i don't run anything untrusted locally (or give untrusted access to anything), but why take a chance

(i run a slightly customized smartos at home)

please tell me this is a remote exploit and I don't have to do anything for my no-users-allowed file servers.

. o O ( la la la I can't hear you )

you mean don't tell you?

from the looks of it, only the SCTP one is something that could be triggered remotely (but probably only if you're running something that's listening for SCTP stuff)

which is probably unlikely

jbk no, you don't need to be running anything that's listening for SCTP

oh that's really unfortunate

... and something seems to be broken with building the latest smartos

hrm... looks like 18043 introduced a flag day I might have missed

(or the 'native' smf build has bugs)

fenix illumos 18043

FEATURE 18043: want libscf smf_XXX_instance variants that take a scf_instance_t (Closed)

↳ illumos.org/issues/18043 | code.illumos.org/c/illumos-gate/+/4683

(and no, that wasn't sarcastic. I'm *really* lucky.)

uugh

the rather byzantine I/O path through sd.c can be annoying