lw: virtio-9p is working great, thanks again. i saw someone mention that with virtio-blk you can pass any /dev storage device to a bhyve guest. haven't tried it yet, but will throw an old drive in tomorrow and see if it works. (github.com/churchers/vm-bhyve/wiki/Virtual-Disks)

Title: Create new page · churchers/vm-bhyve Wiki · GitHub

markmcb: you can add any file or block device as a disk in the VM, but that's not 'passthrough' in the VM sense, i.e. it won't attach the actual device

passthrough is what you can do with NVMe disks for example where the PCI device is connected to the VM so it can access the hardware directly, which is faster

sr-iov

yeah, although i've never managed to get SR-IOV to actually work on FreeBSD, at least with Intel network cards

Asked on Discord yesterday but didn't get anywhere..trying here: fusefs mounts (exfat) as regular user; I get Operation not permitted despite having /dev/fuse and /dev/da* r/w for my user/group. Module is loaded; mounting as root works. I own the mountpoint. I can mount other filesystems (msdosfs, whatnot), just not fusefs. Any idea?

Ltning: what's the exact error?

Just that - "Operation not permitted".

I just realised I was trussing it without -f, which might explain why it didn't give me anything useful. I'll have to try that again.

dmesg say anything?

Nothing in dmesg

Title: Solved - mounting exFAT usb with freebsd | The FreeBSD Forums

that help maybe?

Not really, as mounting as root is no problem

That's all I got buddy, I hope you figure it out.

25892: nmount(0x41726ee3d000,18,0x0) ERR#1 'Operation not permitted'

The mount.exfat (fuse daemon) remains running though

what groups is the user id you are trying to do it in?

operator?

yea, and devfs.rules have /dev/fuse* mode 666

Confirmed by checking the actual entires in /dev

Could it be that I'm unable to create new device nodes? Won't "hidden" nodes (fuse0..9) be created?

I don't know a lot about that...

:(

Nope, that wasn't it

is it a thumb drive?

Yeah

And yes, I have r/w access to /dev/da*

I know at random thumb drives misbehave for me

usually means they are failing

I had a kingston one that I used for windows installer, beat the tar out of it, and one day it just quit.

hi I am trying to get my wacom to work with freebsd, I can see it in usbconfig, and I followed the wiki guide and most recent forum troubleshooting post, but no dice. If I manually try to webcamd -d ugenx.y it, I get webcamd: Cannot find USB device

this is an old wacom and I am pretty sure I got it working in older versions of fbsd

xinput does not list the device

ugen0.2: <Wacom Co.,Ltd. Intuos BT M> at usbus0 as per dmesg

and webcamd -l shows: webcamd [-d ugen0.2] -N Wacom-Co--Ltd--Intuos-BT-M -S 8HH00U2001266 -M 0

does xorg-drivers need to be recompiled with WACOM=on?

noticed after going from 13.2 to 13.3 that box is using WAY more ram. looks like it's zfs but not sure. what changed?

I think my issue is that xorg server does not list the wacom

ok whipped out my backup wacom and instantly recognized

i guess the other one is not supported for reasons I do not understand

noticed after going from 13.2 to 13.3 that box is using WAY more ram. looks like it's zfs but not sure. what changed?

alepzi, No idea. But can you pastebin some data for it? I am interested in what it is doing.

sure, what?

top?

I don't know. You were the one saying it was using more ram. I was interested to know what symptoms were you observing.

I would do a comparison with my systems.

well i have 64gb and with my current workload (few vms open etc) id normally have half of that free. now i have like 4mb

Free RAM is wasted RAM

How are you measuring "free" ram?

I actually have 3% used swap on my laptop, but that's not really a problem. There's lots of laundry and inactive ram (10G total), so it's not really starved

Of course, firefox is eating up the vast majority. about 5-7G, if I were to guess quickly.

i don't use any swap so i can impress the girls in comp class

Hahaha.. That's ... about as smart as not doing backups, though. ;)

That's a solid amount of ARC. Probably because of the I/O related to the VMs, if I were to guess.

alepzi, Example from here: termbin.com/ejqu

It'll shrink quickly once you fire up something that needs physical memory.

My example is running on 13.2-RELEASE.

ya it's just a very noticable diff from 13.2

same exact config and workload

normally i'd have >32gb free

I am totally agreeing with Ltning that free RAM is wasted RAM. It should be put to work in cache.

ZFS got a nice bump 13.2->13.3, unless I'm mistaken. Smarter ARC might be among the changes.

If you look at my 16GB RAM desktop I am only seeing 352M Free but that's a good and happy state here. There is actually a lot of available ram if I fire up something that uses it.

ahhhhh

i bet money regressions are found. i've noticed some disk ops not being as quick

nothing i can repro to file a report on

just a feeling

It might be that you're running without swap so it spends more time doing re-shuffling in memory rather than just nibbling a bit and then carrying on in the background

maybe it was that zfs bug from an optimization that's now turned off?

didn't run swap before

alepzi: No, that was just in very special use cases

ah

Well if you had 32gb free routinely, then it wouldn't make a difference

But now that it's *using* your ram rather than wasting it, having room for some spillover is a good idea

Also if you want to help out, having swap so you can get dumps is also nice :)

ya when 14.1 drops i'm gonna reinstall and set up swap

I might suggest running "top -b -o size" just for giggles.

wtf Xorg size 24G, res 75m

pigdaddy feeding

signal-desktop - size 1116G, res 144M

Got you beaten :D

lmao

thing's obviously coded in visual basic

Bit miffed it didn't switch to T there

Javascript - the VB of the '20's

top also has -o res for resident set size sorting too.

true

Ltning, Numbers starting with 1 need 4 digits of precision.

top -o <cpu|res> -s 1 -SHz -- I can type that while asleep

rwp: I feel cheated.

Think of 1 out of 999 or 1 out of 1001 the same significance but we need 4 digits for the second case.

Yea I know. I still would've taken some satisfaction in seeing a T in there.

Just absolutely batshit that even virtual address spaces grow so large.

Ah... yes, well... (shrug)

In my !work life I obsess over 16 and 32-bit PCs.

(Even @work, but don't tell anyone)

I need to spend some time to study the FreeBSD kernel's memory management. I assume it is doing overcommit these days. I want to learn how to manage it.

Some of us thought it lunacy when we got OS/2 programs that needed more than 512MB virtual address space, and IBM came out with a kernel patch 5 years after OS/2 was supposed to have died that gave us 3.5GB.

That expansion takes things to the general max for 32-bit on intel arch.

But in return we finally got Chromium on OS/2. Which is .. I can't even count the layers of abstraction.

I am sure IBM was getting many requests from customers to make it so they had more address space available.

The main driver, as I understand it, was the need to run VirtualPC and emulate machines with 1+GB RAM to run Windows somethingorother :D

I spent a lot of time doing chip design on HP-UX systems and by default we had 2GB available but if we gave a link option to combine data and stack then we could get to 3GB and so we always did that.

And then I spent a lot of time fixing 32-bit C code to run on 64-bit systems. Which was mostly fixing type bugs. Every new system found bugs for us allowing us to fix them and improve the code base.

But for some reason Java programs I have looked at, mainly hadoop, request a HUGE preallocation of memory right at the start. I mean like needing a gig of memory just to start.

It doesn't use the memory. But it requires it to be available. Of course the kernel usually doesn't actually do anything, yet, but later if the memory is actually used then it might.

But that's an example of where ps/top will show a huge memory size for a process but the resident size is not anywhere near as large.

And that is also a reason that having some swap space available for the kernel to use to figure into the memory calculations is good. The program might ask for a huge allocation right up front, which might not ever be used. Perhaps likely won't ever be used.

In other questions, does anyone know the reason bsd.to has been offline the past few days?

It's been down long enough that it probably justifies changing the /topic to suggest a different one.

wow you designed chips?

i ate a bagful at lunch, thats kinda the same right

alepzi, I did. I worked on our implementation of the JTAG port but mostly worked on cache ram. But then converted over to supporting out CAD/EDA software.

What are you doing walking among us software plebs? ;)

This is PA-RISC stuff, right?

Yes. PA-RISC. But then IA64 Itanium later. Then supporting our distributed system of compute farm mostly for doing simulations.

At one time I had root on 3856 systems in our compute farm. Which I know exactly because I wrote an infrastructure to maintain it.

How would you compare PA-RISC and IA64 from a hardware design perspective? Is it even possible?

With that many systems you don't do it one system at a time. You do it as the entire collective of systems.

Oh I know, i managed 9000 OS/2 workstations for a while. Alone.

PA-RISC is a much more simpler classic RISC type of architecture. IA64 is rather of a huge behemoth due to all of the legacy architecture support needed.

Uh, wat? Legacy architecture? Wasn't the point precisely to break with legacy stuff?

I mean, just for starters, the import a 486 macro cell to be the boot up processor to get the rest of it going! (That was many years ago and I am sure became different after I left.)

No. IA64 for at least the first 7 years I worked on the team there was sold to us internally as being x86 compatible. Which it wasn't. Which created some argument in meetings. But that was the marketing of it.

When I started uni, I sat down in front of a HP-UX machine, the first UNIX I ever met. I felt like a fish on land, but fell in love by the end of the first week. Right next to it was a couple of DEC Alpha/Ultrix machines, and down the hall a bunch of Sun machines.

I mean, compatible means I can buy a PCI-e NIC and plug it in and it will work, right? :-) Well that was not going to be true so it wasn't really compatible.

I typed on a DEC Microvax in university initially and did not appreciate what I was doing at the time. Because it was all new.

Then at HP I worked using the HP Pascal Workstation. Which was an unusual machine. But quickly wiggled my way to get access to our series 500 running HP-UX using an HP 2600 series serial terminal. That was a really run time for me working on that system.

I had the control terminal for a Data General Eclipse as my first "computer" at home. I, like you, did not appreciate what I had been handed; it was older than I was and obsolete already so it felt like trash but it taught me *sooooo* much..

The HP series 500 was the worlds first 32-bit microprocessor. So it was rather of a big deal at the time.

Yea I remember reading about it in some magazine. It was the "adult literature" of choice for me. :D

Then lots and lots of HP-UX for years. Then Debian GNU/Linux for a huge amount of it. And how I am here.

Since I am talking about old things that are fun I spent almost all of my university computer science classes working on paper printing terminals.

DEC Spinwriters. Hazeltines. And so forth. I wrote thousands of lines of code using qed which is a similar sibling to ed.

That's the best way to learn vi by the way. Work with ed for a while. On a paper terminal. Then advance to a CRT with screen addressing.

And of course I am typing this in here with Emacs. Emacs FTW!

But emacs itself is suffering through the thrash that all new stuff goes through. "I love you. You are perfect. Now let's rewrite it all and change everything about you."

And I must run off as real life is calling. Fun trip down memory lane. Chat with you more about it later. TTFN!

so i'm making a scripted bsdinstall for a machine with a jail in it already configured and ready to go. the system distribution dir has my machine's files in it, including the thick jail's base.txz files in /usr/local/jails/containers/testjail/. that "works" and after running the install i can run service jail start testjail. but what's weird is if my installerconfig contains: zfs create -o mountpoint=/usr/local/jails zroot/jails;

zfs create zroot/jails/containers like the handbook says, /usr/local/jails/containers/testjail/ is EMPTY??

so i commented out the zfs stuff from installerconfig, reinstalled and started machine, manually ran the zfs commands, files are still in /usr/local/jails/containers/testjail/, tried service jail start testjail, it fails with mount.devfs: /usr/local/jails/containers/testjail/dev: no such file or dir

ok ya, when the zfs commands are run, the files in /usr/local/jails/containers/* disappear

wtf

jailed property?

sorry what?

"After a dataset is attached to a jail and the jailed property is set, a jailed file system cannot be mounted outside the jail, since the jail administrator might have set the mount point to an unacceptable value." zfs-jail(8)

i guess what i gotta do is let the installerconfig run the zfs commands, and THEN move the files into the path from a tmp dir

weird. so in the installerconfig i run the zfs commands. then AFTER that i mv /tmp/testjail /usr/local/jails/containers/ but when i reboot, there's nothing in that dir

zfs create -o mountpoint=/usr/local/jails zroot/jails; zfs create zroot/jails/containers; that doesn't create the /usr/local/jails and /usr/local/jails/containers dirs?

ok when i run those zfs commands after installation manually, it creates the dirs. but when the commands are in installerconfig, the dirs don't get created. is that a bug or?

if you're in an installation environment, it could be that the parent directories are on a readonly filesystem

how can i check that?

interactively you can check the output of `mount`

alepzi: the "mount" command

damn, I'm too slow

k sec

i've never done installation automation, so i don't know how to reason about your overall setup, but the default installer involves readonly mount of /

(at least, as far as i remember)

k got it. what should be looking for in the mount output

ro or rw -- for read only and read write

ok got theinfo...

there's a bunch, which line?

btw zroot get altroot says /mnt local

alepzi, If you are wondering what zfs zpool commands were actually getting run then run "zpool history" and look to see what the bsdinstaller did when it did it.

My first thought after reading your description is that you unpacked the files into a directory and then created a dataset on top of the directory of unpacked files. That will mount the dataset /on top of/ the directory hiding any contents.