hello

hi

Anyone got any tips on optimizing NVMe on 13.x?

I got an NVMe drive last year and it's been fine. Doing some I/O heavy work right now and wondering if there's anything I can/should tune for it. No big deal, but it feels like it's slower than it should be.

_xor: tunefs -t enable if it's ufs ?

So.. My workstation is currently running 13.2-BETA2. I'd like to upgrade it to -RELEASE. How do I migrate from -dbg components?

Or does it matter that I'm installing, among other things, kernel/generic-dbg ?

ghoti: -dbg pkg is just the symbols for the kernel, you should be able to just upgrade these as normal

anyone have a recommendation for the source of the most in-depth and complete info on kernel config options that's not "use the source, newb"?

did you read NOTES ?

RhodiumToad, yep

wait, maybe i don't know what I need

I checked the NOTES, LINT, and GENERIC files for my arch, combined the includes, looked around, and couldn't find good info on the options for CPU

I had to go to the handbook, find the info on config, search config manpages, check the source, etc.

but I see now that's only valid for "cpu" since other values are just the name itself

I couldn't find WITH_CTF anywhere until i did a web search either

what I hoped to find is something with all the options and the possible values for those options if applicable

if it also included which options are related, that would be epic

for WITH_/WITHOUT_ knobs, you want src.conf(5), though the details can be pretty sparse at times

WITH_CTF is a kernel option not a src.conf option

kevans, checking. that's exaclty my problem. LINT has a comment that's the expanded name of the option. No real explanation

for example, what exaclty is "options HID_DEBUG # enable debug msgs"? am I debugging the HID drivers? am I debugging the HID devices? am I adding debug symbols to the HID drivers or debug messages enabled or other debug functions? JTAG support for the I2C HIDs? WTF?

(more precisely, it goes in "makeoptions" in the kernel config)

_DEBUG options generally mean verbose messages from the relevant driver or module

RhodiumToad, sure, I was just hoping for better docs for each option. Would anyone support starting a project that's compatible with the current state, but extends the interface for adding config options to support adding it's current name, value, description, supported values, data type, extended description, link to documentation, and link to project?

The idea being that we slowly convert each entry until the kernel has the info to auto-generate either a Handbook-style documentation set or even a config file directly?

right, but most of the kernel WITH* knobs are mirrored versions of the src.conf(5) knobs intentionally

I don't really want to change the kernel source so this would have to be metadata and external to the kernel project

not that they're marked +kernel or not

checkign src.conf now

for stuff like HID_DEBUG, there's likely no good documentation

those are config(8) options

kevans, that's exactly why I'm suggesting a metadata parser that we can run over the kernel and auto-generate both a Handbook-style documentation collection and a config file direclty

in an ideal world, in the version of config(8) that we haven't written yet, we provide a description in the option metadata that's naturally machine-readable without tricks

kevans, maybe that's a natural fit for the thing I'm talking about

I see src.conf has some stuff, but not exaclty a full set or exmaple values

right, figuring out everything that can be tuned is tricky; most of the kernel stuff you want is defined in sys/conf/options and documented (if not there, and if at all) in the various LINT configs

there are some WITH* knobs that are defined in sys/conf/kern.opts.mk, many of these can generically be interpreted as "equivalent to whatever src.conf(5) has to say about them, if anything"

okay, let's say I was going to do the heavy lifting on this. Who would I ask about making sure I don't start something that won't be included because it somehow affects or upsets the authors, contributors, sponsors, or users and is only helpful if needed and unobtrusive if not

I checked sys/conf/options

I see an include for an opt_*.h file and those files aren't available in the tree. What do I do to find the next connection there?

opt_* file contents are derived from sys/conf/options* (it's the second field on the content lines, "which header this goes in when it's defined") based on what's enabled or not enabled in the kernel config

opt_* are autogenerated

The reason I ask is because I'm a decent programmer, I've got some spare time, and I can go through the source and track down each config option and get a grip on how the values are used, and then generate a quick explanation. I wanted to try to enlist some AI help. I know that it's possible without getting in anyone's way. I just don't want to do a lot of this work to see it go to waste immediately.

(well, that's probably phrased better the other way around, but the idea is that if you want to see what might come from this header, sys/conf/options* are where you answer this)

okay, they're just #ifdef's in source files

I thought it was some make file magic, but nope

so, presumably, all kerenel #define's and #ifdef's are potential config options for the kernel?

is there a code style guideline for the kernel that bans use of #define's outside the config system?

negative

you're likely better off to approach it from the other way; take a knob out of sys/conf/options*, grep around for where it's used to get a feel for what it touches

I would probably pitch the idea of thoroughly documenting this stuff to freebsd-arch@, btw, with some specifics of how you want the end result to look

mailign list?

yeah

okay, yeah, here's my plan: I'm going to take one option and follow through on a the minimum usable implementation to get a flow going. then i'm going to think about what options would be helpful or worth investigating. then i'll take that to freebsd-arch@

I don't want to run in there and just dump this and then have no follow-up ready

this is a major task, so it's definitely worth getting some concrete buy-in... we seem to have 900+ config(8) options across all architectures

most of them probably not worth documenting, mind you

kevans, do you really think it's that big? I figured I could get GENERIC for amd64 done in a couple of weekends if I design the process correctly

it depends on how you're scoping it; obviously the vast majority of those likely aren't used in any GENERIC config

but they do exist

right now, I'm thinking: get GENERIC, grep options into one file, device into another, for each option create dir in name, grep all files with '#define OPTION" and add filenames to a files.txt, for each file, grep -B 100 -A 100, write a script to find the right #endif and the previous comment, write to "source_X.txt" for code and "commment_X.txt" for comments where X is the pathless filename disambiguated

then feed all that to a GPT/LLM/NLP chatbot. results would be: option name, option values, option descrition short, option description long or URL of project or related source files. that's off the top of my head. I'm sure I can do better

maybe I'm wasting time

okay, maybe I'm thinking about this from the wrong perspective. Can anyone help me find the right audience for this question? I suppose kernel developers vs. users who configure kernels but don't spend time with the project open in an IDE and aren't ready to chase down each #define..

what are you talking about man?

what are you trying to accomplish?

document kernel config options as quickly and automatically as possible without impacting or upsetting kernel dev's and in a way that's useful to kernel config users

just how often are you building kernels that require these distinct options?

given that gpt just makes shit up, i don't think it's properly suited for documenting techical aspects of the freebsd kernel

rtprio, I would like to make something that others use because I plan on customizing kernels for at least 4 machines and I expect to have to build/configure several dozen times, so if I can get my task done faster and help others, I'd like to at least try

rtprio, I agree that GPT would have to be filtered through a, hopefully competent, developer's eye.

did you like, you know, read NOTES for more info on those options?

also, once you pare it down a lot you can load modules so you don't to customize that much

rtprio, yep, LINT->NOTES->Handbook->man pages->config files->more man pages->source->here

that's what got me thinkign this could be done better

then just send patches for NOTES why the fuck would anyone try and make this so complicated

rtprio, because it's so worthless right now that NOTES might as well be called NONSENSE.

how do you figure

let me see, what does NOTE say about "cardbus" for example

# cardbus: CardBus slots

aka pcmica

oh, well great, now I know whether that's important to include in my kernel

if you don't know, you probably don't have it

wouldn't have guessed that if it was the last question on who wants to be a millionaire

rtprio, I have a PCMCIA slot. never used it but also never heard of it being called CardBus

cardbus is technically the extension of pcmcia to 32 bits

and I used PCMCIA Wifi, TV tuner, GSM data, and GPS and had half of that working in Linux. I probably never did in FreeBSD because WTF IS CARDBUS?

can we just agree that the kernel options docs are shit? I don't see a reason to argue on taht point. everyone knows but not everyone thinks that a full GUI menu is necessary. I don't like that myself and wouldn't waste time on it. But some docs??

well, i don't think that's the fault of the documentation

include cardbus if you need cardbus but it's on you to know what it is

just like all the shit in that list

do you want to copypasta whatever gpt comes up with to document cardbus for the 1% of people who might be using

so I should know alternative names for every single thing in my device tree and all the options and values I need to use instead of having that available and updated from the source? why don't I just write the whole OS myself? not much of a step if I only implement the options I need

no need to get huffy because your idea is not a good one

also, as i tried to say, you don't need it in your kernel, it will load as a module if it's needed

rtprio, I specifically want to use GPT to take the source code #ifdef'd in the config and explain in English how it would alter the behavior of FreeBSD. I think it can do a decent job that I can review.

ok look at the code and tell me if you can better explain what cardbus is

that's ridiculous

rtprio, I apologize if you misread my mood as "huffy" but please don't confuse the fact that you lack a usecase for this with the usefulness of the idea in others' hands

right. i've read NOTES and gotten by for 15+ years. as most the people in this channel. the ones who could be bothered to build a kernel

rtprio, I am trying to make something better than NOTES that explains what "device cardbus" does better than it's currenlty done in NOTES. If that's not useful to you, and you aren't obstructed by this, go tell someone else about your opionion. I want to help some and impede none and their feedback mattes.

ok, well i look forward to see what you come up with

wait, maybe I'm wrong

i mean, perhaps i last googled 'cardbus' like 10 years ago, well after i threw out all my hardware that might have had cardbus

your feedback might matter if you have anything to say about why this would be entirely useless in your experience.

okay, forget cardbus. what about this:

options SC_DEBUG_LEVEL=5 # Syscons debug level

what is that? what does 5 mean? is 0 off or the first option?

# Yet more undocumented options for linting.

options MAXFILES=999

what does this do?

ok, you got me on that one

device speaker #Play IBM BASIC-style noises out your speaker

but unless youre debugging syscons, there's no reason why you'd ever give a shit

do you not know what a pc speaker is?

which speakers? the builtin? my tv's? my neighbors? the PC speaker connected to the PIT? do those still exist?

PIT?

yes, the header on the motherboard, from olden times

i know what a PC speaker is. does it say "PC speaker" anywhere?

it says 'speaker' i feel like that's close enough

not 'sound card' or whatever

just why are you compiling kernels anyway

do you have weird hardware that's not in GENERIC

options ATKBD_DFLT_KEYMAP # specify the built-in keymap

makeoptions ATKBD_DFLT_KEYMAP=fr.dvorak

why is this using options to define and make options to set values? how would I know why it does that?

the keymap is a separate file that has to be processed by the makefile at some stage

so it can't just be a #define

ok

does the system with PCMICA still have an AT keyboard?

you're killing me, smalls

what's wrong with AT keyboards?

# More undocumented options for linting. Note that documenting these are not considered an affront.

options KBDIO_DEBUG=2,options KBD_MAXRETRY=4,options KBD_MAXWAIT=6,options KBD_RESETDELAY=201,options PSM_DEBUG=1,options TIMER_FREQ=((14318182+6)/12),options VM_KMEM_SIZE,options VM_KMEM_SIZE_MAX,options VM_KMEM_SIZE_SCALE

personally I find it annoying that the keyboard map now has to be specified in 4 places (ATKBD, KBDMUX, HKBD, UKBD)

nothing's wrong with it

RhodiumToad, this would be use case for documentation linking related options. thanks for the heads up

anyway, if anyone has ideas on how to amke this better, ways to be unobstructive to dev work, or reasons for me to reconsider doing this, I'd love to hear your thoughts

ty freebsd for being sane and supporting "df -g"

Demosthenex: wait until you realize freebsd's ls doesn't have --group-directories-first

armin: alas neither linux nor freebsd support grep -p for paragraphs.

I think you can get the same result (as df -g) under Linux by using the BLOCKSIZE variable

brb

Hi, I got this problem: i carefully followed the instructions of installing the XFCE desktop environment and the LightDM display manager, but after a reboot, the DM won't start.

hi, do you know how can I install both, terminal and gtk3 version of vim?

every command was executed correctly and I doule-checked the writing in /etc/fstab

having some trouble on understanding an issue I have. trying to use wildcards to perform actions on very large list of files =)

example on error msg: exec: Failed to execute process '/usr/bin/grep': the total size of the argument list and exported variables (883kB) exceeds the OS limit of 512kB.

how do I change this OS limit?

I also seem to hit some limit when I try to browse the same directory with thunar.. =)

perhaps the issues I have is related - but I seem to understand the problem to little to even know where to start googling it =D

You probably need to pipe to xargs.

Hello there, got a node process using "21.3G" of "VIRT" ram, any clue on that? i.ibb.co/nBCj2S9/highmem.png

tercaL: any idea what it's doing?

tercaL: the clue is that JavaScript is a terrible thing to begin with

wouldn't it be if node had introspection where you could find out what it's doing and why so much memory is bound up

And by the way, the Swp: shows 0K/0K, but the process uses that much amount.. How can this be?

Please use ps(1) - whatever that tool is, it's probably not reporting the right values.

ps(1) understands the difference between virtual size and resident set size.

I've seen JavaScript stuff take up over 1TB of virtual memory all by itself, while being well under 1GB of resident set size - so it's not exactly unimportant.

well, that screenshot says RES 220M

Interesting and really cool details, thank you all!

Will dig further into the proc. itself..

is it possible to change this limit: the total size of the argument list and exported variables (883kB) exceeds the OS limit of 512kB

looking at the sysctl list, but not sure which could be relarted

kern.argmax maybe?

miltux: seems to make no difference in boot/loader.conf and read only with sysctl

I'm trying to do use linux emulation, the application works but there is no sound, do I need to setup something extra to make sound be available through linux emu?

drobban: it's probably this one, but no idea how to go about changing it

miltux: thanks. I tried, it seems like I dont either =D

wrt pkg: . wrong user or group ownership (expected 0/0 versus actual 1002/0)

during `pkg --rootdir ... install` this is a reasonable error as this is all as non-root

is it possible to tell pkg to just go ahead anyway?

drobban: If you really, really need it, you have to change /usr/src/sys/sys/syslimits.h and compile a new kernel, I guess. Search for ARG_MAX

oh we have a #pkg channel I will ask there

miltux: =) what I was affraid of.

miltux: thx!

drobban: np

dch: how many files are we talking about? If they're not that many, perhaps @owner and @mode in plist file would be enough

miltux: this is using pkg install as a non-root user, not modifying a port itself

dch: Ah! sorry

An in-line IP change question: I've a jail configuration like; web { ip4.addr = lo1|10.10.10.2; jid = 1; } and so far, it works fine. I'd like to change its ip within the root terminal in host, with the command; jail -m name=web ip4.addr=10.10.10.44, and this works too it seems, because "jls" reports the updated IP. However, when I enter into jail, it seems it didn't change the IP address. And when I restart the jail, the older IP (from jail.conf) seems

back. Any solution to this?

Is there any command like to apply such changes to the jail, without restarting it?

maybe jexec the proper ifconfig command to change the IP?

jschmidt3786: we now have ifconfig -j !

i imported a disk image that was running zfs into a different system; and files i was expecting in /var are not there

there's only "audit crash log mail tmp" but /var/puppet is suspiciously missing

where are the other directories that used to live in /var

maybe a different dataset that you didn't import/mount or maybe a different dataset on a different pool that you didn't import. you'd have to look.

that's the only pool

i've ran into this before and unfortunatly cannot remember the explanation

narrowed the failure mode on PR 273372. IOV for bhyve vms works if i reduce the vfs (or possibly if all vfs are homogenous passthrough). will test more later to see if i can narrow it down further.

273372 – SR-IOV Networking in Bhyve Causes Chelsio T520-SO-CR to Fail on Host, Kernel Panic if Reset bugs.freebsd.org/bugzilla/show_bug.cgi?id=273372

meena: gotcha, I forgot about that.

i think it has to do with `canmount`

Hi! Is there a way to list all volumes inside a ZFS filesystem?

volumes are inside pool, same as fs

thanks!

if you mean the paths, zfs list -t volume -r path/to/dataset

yuripv: yes, this is perfect! Thanks!

sorry if my first reply didn't make sense, i just read your question wrong

yuripv: np! thanks you very much for your help :)

hi, anyone knows where i can get 15.0-CURRENT from?

link is dead -> download.freebsd.org/snapshots/amd64/amd64/ISO-IMAGES/15.0

opnsense isn’t FreeBSD based?

It is stated that it is based on FreeBSD.

public-static: check back in a day or two

will do

(the first builds post-stable/14 failed, iirc because packages weren't quite there yet)

public-static: found mirror for i386 ftp2.de.freebsd.org/pub/FreeBSD/snapshots/ISO-IMAGES/15.0

Title: Index of /pub/FreeBSD/snapshots/ISO-IMAGES/15.0/

Hello everyone, I am having a problem getting a WireGuard server (FreeBSD 13.2) to have good performance when connecting to other devices on the same LAN as the server itself. For example, if I am at work, and the work server is connected to my WG server at my datacenter, the work server has a hard time with VNC to a Windows VM on that same network, or a file server on the same network. I'm not sure how to diagnose this as I'm not the most

fluent in firewall/IPFW/PF (for NAT) configurations.

for example, my VNC speeds are at 127Kbit/s whereas without WG, the speeds hover aroung 80-110 Mbit/s

Hey, could anyone help me getting my USB controller to work without root in Dolphin-emu?

I can setup any udev stuff myself.. i think

but I tried messing with permissions around /dev/usbXXX, i was getting "Other error" from libusb

are these the wrong dev files? is there a specific tool for changing usb device permissions?

Of course. These work fine in Root

Controller works as is.

ugen1.2: <Nintendo WUP-028> at usbus1

uhid1 on uhub1

uhid1: <Nintendo WUP-028, class 0/0, rev 2.00/1.00, addr 1> on usbus1

#chmod 777 /dev/usb/1.2.0 causes the libusb error

running it as root works fine.

at 3am what runs that spins hdd so much? its not scrub but what else?

got to do with setuid ?

Beladona, looks like "periodic daily". See periodic(8).

V_PauAmma_V ya but what in it?

something strange has happened when I did the 13.2p3 update

with the boot environments

check it out:

FBSD-13.1-p2 NR / 29.6G 2022-09-14 08:03

but

root@daneel:~ # uname -a

FreeBSD daneel 13.2-RELEASE-p3 FreeBSD 13.2-RELEASE-p3 GENERIC amd64

root@daneel:~ # freebsd-version -kru

13.2-RELEASE-p3

13.2-RELEASE-p3

13.2-RELEASE-p3

that's really whacked out, that's a very old boot environment I had back in 2022,

it somehow, it put the new p3 update into that

and put it into now/running mode

what do you make of that?

excuse me, the next/running mode

I just used the standard freebsd-update fetch install

and rebooted

yeah I'm definitely not insane

zroot/ROOT/FBSD-13.1-p2 mounted yes -

any ideas? I cannot of any reason this should have happened

I guess I can just rename the boot environment, but this seems crazy

DrKK`: is freebsd-update.conf still unchanged? does it enable boot envs?

I'll check that

but I noticed my "default" is gone

I don't have a "default" BE

let's see if it's operator error

hold one

Beladona, see the manual page. It will tell you where under /etc/periodic to look to see what all is done. (Although from your mention of setuid, I'd guess 450.status-security, which in turn runs "periodic security", and through that 100.chksetuid. But that's only a guess.)

# Create a new boot environment when installing patches

# CreateBootEnv yes

so that should be fine

it's probably my dumb ass somehow deleted the default BE

let me check the zfs history

2022-02-01.23:43:39 zfs destroy -r zroot/ROOT/default

^^^

whoops

so then something else became my "default"

and I didn't notice until just now, somehow, after 1.5 years

that seems hard to believe. But. OK.

cancel inquiry. I'll assume everything is fine unless this happens again at the next point upgrade

is there an sctp equivalent for unix domain sockets?

do sctp streams eat up file descriptors?