channel logs for 2004 - 2010 are archived at http://tunes.org/~nef/logs/old/ ·· can't be searched

#osdev2 = #osdev @ Libera from 23may2021 to present

#osdev @ OPN/FreeNode from 3apr2001 to 23may2021

all other channels are on OPN/FreeNode from 2004 to present

http://bespin.org/~qz/search/?view=1&c=osdev2&y=23&m=1&d=23

Monday, 23 January 2023

06:04:00 <klys> what's a pcie-gpp and where can I find out about its protocol
09:21:00 <wikan> hi :)
09:21:00 * wikan waves
09:21:00 <wikan> can you tell me how important is advanced math while you want to write your own os?
09:22:00 <zid`> 0
09:22:00 <EthicsGradient> you only need understand two numbers: 0 and 1 :^)
09:22:00 <wikan> why such many?
09:23:00 <hmmmm> very important
09:23:00 <wikan> it is extremely hard to remember them :)
09:23:00 <hmmmm> you need to understand the quaternions in order to successfully create an OS
09:23:00 <wikan> i am working on converting numbers from string now
09:23:00 <hmmmm> you are writing your own libc?
09:23:00 <zid`> hmmmm: Don't forget 18 dimensional topology
09:24:00 <wikan> my math is so poor I always need debug my math code :|
09:24:00 <hmmmm> mmm yes algebraic topology in general you need
09:24:00 <wikan> no, i am writing library for my tools. Console argument management library.
09:25:00 <wikan> you type: SOMETHINK=:0 ./myapp --argument bla bla -Xla -DBLA=xxx
09:25:00 <gog> advanced math? not really
09:25:00 <wikan> and library takes care of it
09:25:00 <gog> but understanding algebra is pretty critical to programming in general since programming is essentailly algebraic
09:25:00 <wikan> i only connect callbacks without checking env and args manualy
09:26:00 <wikan> i like to say "my brain is database like brain not math brain"
09:26:00 <hmmmm> how will you implement SSL without Galios theory?
09:26:00 <hmmmm> back to school with you.
09:26:00 <gog> true
09:26:00 <wikan> it is easy to me to learn information but when I need to do math, I can't imagine it most times
09:27:00 <wikan> well I hope there will be people whom will like my os and will write math code :D
09:27:00 <zid`> gog: 18 dimensional topology vital idk what you're talking about
09:28:00 <sham1> Cryptography and finite state automata
09:28:00 <hmmmm> category theory and lambda calculus are essential.
09:28:00 <hmmmm> (for certain programming languages)
09:29:00 <sham1> The former also being known as "abstract nonsense"
09:29:00 <wikan> but for micro kernel?
09:30:00 <wikan> memory management, dist operations,etc? how about it
09:31:00 <sham1> Topology should be useful for things like memory management
09:31:00 <wikan> i like to watch Andreas Kling youtube channel (the guy whom wrote SerenityOS)
09:31:00 <wikan> amazing person
09:31:00 <sham1> Because you'd want to find ways to split memory into fair divisions
09:32:00 <wikan> topology is a mathematical thing? I thought it releated to cartography
09:32:00 <sham1> Yeah. I suppose you were thinking of topography
09:32:00 <wikan> yea, right
09:32:00 <hmmmm> topology is a more generalized version of analysis.
09:33:00 <wikan> so... my memory managment will be not fair :D
09:33:00 <sham1> Well topology will be more useful in proving things about fair memory allocation
09:33:00 <sham1> I.e. that it's possible
09:33:00 <EthicsGradient> sham1: but also how do you define "fair" ?
09:34:00 <wikan> "to insall this OS you need at least 16B of RAM" and problem solved :)
09:35:00 <sham1> EthicsGradient: If you have two processes, you divide the allocation evenly. Another would be scheduling, you'd want to give enough quanta for both and if they're the same priority, you'd want to give both of them the same amount of time
09:35:00 <EthicsGradient> yeah makes sense
09:36:00 <zid`> does anybody do binpacking for process allocation across cores
09:37:00 <sham1> Well you'd probably arrive at it eventually trying to make it so that your cores are equally occupied
09:38:00 <sham1> Moving one thread to another core when it's idle and whatnot to balance the load
09:40:00 <zid`> binpacking and 'fiddling a lot' are approximately the same thing anyway :P
09:45:00 <sham1> Of course a big part of it becomes "how many threads should be in a core"
09:47:00 <moon-child> zid`: probably not
09:47:00 <gog> how do you solve a problem like scheduling? how do you catch a thread and pin it to a core
09:47:00 <moon-child> trying to perfectly balance work is hard to do when you're a kernel because: 1) you don't have enough knowledge about the application; 2) the landscape is constantly changing; 3) context switching is expensive
09:48:00 <moon-child> there are libs like starpu that try to do a perfect job of scheduling in userspace, avoiding those pitfalls
09:48:00 <moon-child> (or, at least, heuristic, since optimal is np iirc)
09:48:00 <moon-child> (and yes starpu is the dumbest fucking name ever)
09:49:00 <moon-child> gog: very carefully
09:50:00 <sham1> Obviously what you do is that you move the scheduler to userspace and have it handle the policy. After all, context switches are expensive anyway, so might as well /s
09:51:00 <gog> yes
09:51:00 <gog> switch context 4 times instead of 2 times
09:51:00 <gog> always be switching context
09:51:00 <sham1> Picokernels
09:52:00 <moon-child> sham1: high-performance systems are all one thread per core and handling the multiplexing of work over cores--whether you call that scheduling or something else--themselves
09:53:00 <sham1> Ah, so co-operative
09:53:00 <moon-child> and two-level schedulers are popular (erlang, go, java) for the same reason
09:54:00 <moon-child> (although I would rather have just the one scheduler and have it be fast; ditch the hardware-memory-protected kernel)
09:55:00 * wikan waves
09:55:00 <moon-child> wikan: wassup
09:55:00 <wikan> <^*.*^>/
09:56:00 <wikan> i will strart writing my os when I get information I don't have cancer anymore
09:58:00 <moon-child> best of luck
09:58:00 <wikan> probably it is going well
09:58:00 <wikan> cancer markers dropped down from 9k to about 20 ;)
09:59:00 <gog> nice
10:00:00 <wikan> now i am working on some tools because I want to write my own os in my own language
10:01:00 <moon-child> what sort of language?
10:01:00 <wikan> it looks like mix between c and vim script :D
10:01:00 <sham1> Eugfh
10:01:00 <moon-child> well. Best of luck with that
10:01:00 <sham1> I'm not sure VimL is the thing to take inspiration from
10:02:00 <wikan> well I just use prefixes
10:02:00 <sham1> Okay, that's better
10:02:00 <wikan> i like this idea becuase it is clear for me
10:03:00 <wikan> there is no strlen(), but func:@com/std:strlen()
10:03:00 <wikan> and you know where this function is
10:04:00 <wikan> you can't create variable like "variable" you must prefix it "var:name"
10:04:00 <wikan> or "const:@com/std:INT32_MAX"
10:05:00 <wikan> for me it is easy to follow
10:07:00 <wikan> and @com is undefined. You define it when you call compiler. It is namespace. You define @com is a "this directory"
10:07:00 <wikan> so easy to work with subprojects like libs, etc
10:09:00 <wikan> for os it would looks like: @drivers:fs/ntfs:init()
10:09:00 <wikan> wrong, @drivers/fs/ntsf:init()
10:10:00 <zid`> are you talking to someone
10:11:00 <wikan> i know i know
10:11:00 <wikan> i am excited ;)
10:55:00 <heat> kernel
10:58:00 <moon-child> cold
11:00:00 <heat> heat
11:01:00 <hmmmm> warmth
11:01:00 * wikan says bye
11:01:00 * wikan wawes
11:01:00 * wikan waves
11:03:00 <heat> hmmmmmm
11:14:00 <Ermine> uncold
11:14:00 <Ermine> or, coldn't
11:18:00 <FireFly> cold's
11:21:00 <mjg> gift.unwrap()?
11:42:00 <epony> try to use half-words that would make better virolunar poetry in hieroglyphs, use pictographs and make bottom to top for more fun reading
11:42:00 <heat> mjg, btw phoronix tested booting BSDs on 13th gen Intel
11:42:00 <heat> the only one that booted was OpenBSD :v
11:42:00 <moon-child> wer sapphire rapids
11:43:00 <mjg> heat: :p
11:43:00 <epony> OpenBSD has its own ACPI implementation
11:43:00 <mjg> heat: i tested that at work, booted fine, but perf was ass
11:43:00 <epony> ass is sometimes perf too
11:43:00 <mjg> heat: due to lack of understanding of the topology
11:43:00 <moon-child> because of big.little stuff?
11:43:00 <mjg> yea
11:43:00 <moon-child> isn't it the same topology as 12th gen?
11:43:00 <epony> no, because it's arm in disguise
11:43:00 <epony> and has eff and perf cores separation
11:43:00 <heat> i firmly believe literally no one uses FreeBSD
11:44:00 <heat> hell, I bet it doesn't compile
11:44:00 <moon-child> also I thought I saw a patch somewhere that dealt with that
11:44:00 <epony> FreeBSD is used a lot
11:44:00 <moon-child> heat: no one uses freebsd on desktop
11:44:00 <moon-child> (except me apparently)
11:44:00 <epony> but not that much on laptops
11:44:00 <mjg> heat: i confirm, tried to compile freebsd yesterday!
11:44:00 <moon-child> and no one uses big.little on servers
11:44:00 <mjg> nobody has numa on servers
11:44:00 <epony> FreeBSD is used on desktops a lot
11:44:00 <epony> but nobod uses arm on servers
11:44:00 <epony> because it's useless
11:45:00 <epony> I mean, the most that gets usage is Sparc and such that are RISC
11:45:00 <epony> and Power / PPC
11:45:00 <epony> and even HPPA
11:46:00 <epony> so RISC has place on the servers and high performance computers, only Arm cores are small, too dense, and overheat and are not suitable for high throughput machines as is so far
11:46:00 <epony> plus the lack of SBSA support and internal buses in the earlier gens was a no-go
11:46:00 <heat> https://www.phoronix.com/news/BSDs-On-Intel-Raptor-Lake
11:46:00 <bslsk05> ​www.phoronix.com: Trying Out The BSDs On The Intel Core i9 13900K "Raptor Lake" - Phoronix
11:47:00 <epony> the performance on OpenBSD is not great at all because it's SMP/SMT contended and has kernel space locking
11:48:00 <mjg> heat: well that is a little different than whaty ou led me to believe
11:48:00 <epony> and is like FreeBDS before the giant lock removal from the kernel, so think 2002-2004 and similar
11:48:00 <heat> why
11:48:00 <moon-child> kinda throws into question the whole 'we start up in real mode for backwards compat reasons'
11:48:00 <mjg> heat: the kernel finishing boot but not able to find the boot dev is very different than straight up not booting
11:48:00 <heat> netbsd actually traps
11:48:00 <mjg> heat: in comparison dfly panicked on something
11:48:00 <epony> FreeBSD has the lock removed and is not SMP/SMT contended since 2004-2008 (even before Linux got SMP unlocked)
11:49:00 <heat> no, dragonfly booted to a cmdline, freebsd got stuck somewhere, net died, OpenBSD is the GOAT
11:49:00 <epony> but the perf/eff cores shuffling inside the arm-like new gen CPUs fro Intel is going to suck on Windows and Linux too
11:49:00 <mjg> oh sorry, the panic is netbsd
11:50:00 <epony> that's the ACPI
11:50:00 <epony> tweaking it a bit on the others will fix it
11:50:00 <mjg> heat: oh hehe
11:50:00 <epony> but the pefrormance is going to suck on all systems which do not have "changes" for the context switching and migration of processes between the cores
11:50:00 <mjg> heat: did the guy try installing openbsd though?
11:50:00 <mjg> heat: afair they boot with a ramdisk
11:51:00 <mjg> heat: meaning by the stage he reached in the screenshot there is no i/o attempts by the kernel
11:51:00 <heat> But then when proceeding with the installer, the wired and wireless networking for this ASUS Z790 motherboard were not working on this latest OpenBSD release. Thus resorting to an older USB network adapter to have a network connection
11:51:00 <mjg> heat: for all ik now it would have failed the same way fbsd did
11:51:00 <epony> and OpenBSD has support for both DT and AHCI for the device enumeration
11:51:00 <heat> but it seemed like it does install
11:51:00 <heat> "But as for how OpenBSD on Raptor Lake compares to Linux for performance, that's saved for another article"
11:51:00 <epony> so technically should work either way on ARM too, in contrast to the other BSDs
11:51:00 <mjg> ok, that i'm not gonna suffer
11:51:00 <heat> can't wait for OpenBSD vs Linux
11:52:00 <epony> no Linux has better performance than OpenBSD because Linux is SMP-unlocked
11:52:00 <mjg> i don't know if i can get hadns on raptor lake easily
11:52:00 <epony> and has more support in drivers and NDAs etc
11:52:00 <mjg> everything has better performance than openbsd, evne onyx
11:52:00 <epony> no
11:52:00 <mjg> ask heat
11:52:00 <sham1> Year of the FreeBSD desktop shall be the next epoch once the YOTLD happens
11:52:00 <mjg> dude we benchmarked some of it
11:52:00 <heat> the only BSD I don't beat is Free
11:52:00 <epony> plan9 has worse performance than OpenBSD
11:53:00 <heat> but that's a moot point because FreeBSD doesn't even boot :v
11:53:00 <epony> FreeBSD is the most interesting of the BSDs and OpenBSD is the most useful
11:53:00 <mjg> heat: you hurt my feelings
11:53:00 <epony> NetBSD is like OpenBSD with deficits
11:53:00 * mjg is going to shit some on rust to chilll down
11:53:00 <sham1> NetBSD is like OpenBSD without the hilarity that is Theo sometimes
11:53:00 <mjg> what
11:54:00 <epony> the advantage of NetBSD is porting and drivers, but it cross compiles and some platforms have not been booted / tested on real HW for decades
11:54:00 <mjg> sham1: i recommend you google for posts by maxv on netbsd lists
11:54:00 <mjg> sham1: really :)
11:54:00 <sham1> Ooh, well then
11:54:00 <epony> hilarity is when NetBSD tries to claim performance and FreeBSD shows up
11:54:00 <heat> mjg, sounds like something is deeply wrong with BSD drivers such that they cannot boot a new gen at all
11:54:00 <heat> do they need a list of PCI ids for these devices?
11:54:00 <mjg> heat: i may or may not be able to get some hands on a raptor box and we will see
11:55:00 <mjg> heat: at $work we got ice lake
11:55:00 <epony> the NetBSD lists are hard to Google.. because there are too few subscribers ;-)
11:55:00 <mjg> heat: well you do need to add IDs for drivers to attach
11:55:00 <epony> the new Intel gen is shit
11:55:00 <mjg> heat: it may be for a lot of it adding the Ids is literally all that's needed
11:55:00 <epony> use Amd
11:56:00 <sham1> Well that's just Intel. I wonder how AMD is doing
11:56:00 <heat> but this is standard hw
11:56:00 <heat> you can bind based on the class/subclass/progif ID in PCI
11:56:00 <sham1> I want a threadripper. They're sadly too expensive
11:56:00 <epony> and stay away from APUs 'cause the LinuxDRM is like sucking and blowing assballoons at the same time
11:57:00 <heat> mjg, if you need to add EVERY PCI VID/DID PAIR EVER for a device, your device binding sucks
11:57:00 <epony> there is nothing to justify a Turdreaper.. software is lagging and the kernels are contended for parallelism
11:58:00 <epony> it's probably only targeted by some workstation commercial crapware OSes and middleware / hyperviruses (visors)
11:58:00 <heat> and it's not like xhci or ahci or nvme actually changed lately. so what's the issue? unless you need major quirks for some broken hw, things should work
11:59:00 <epony> yeah, you're getting your performance fix from IBM
11:59:00 <epony> that's the conclusion
11:59:00 <epony> cause Linux alone is not there too "as is"
11:59:00 <sham1> No one was ever fired for buying Big Blue
12:00:00 <epony> except IBM is not used in data centres
12:00:00 <mjg> heat: i'm not saying it does not
12:00:00 <mjg> heat: .. suck
12:01:00 <epony> all of the big tech clouds are custom tweaks consumer mainboards and server variants for the infrastructure/db/application servers
12:01:00 <mjg> heat: looks like amazomn only has cpus up to ice lake
12:01:00 <mjg> notgud
12:02:00 <epony> only laptops get the latest gens typically
12:02:00 <epony> and then desktoys
12:02:00 <epony> and last servers
12:02:00 <epony> so much as to servers skip entire gens and cycles
12:03:00 <epony> so the frequency of releases of new models is highest with Apple / Arm / nVidia and consumer / mobile
12:03:00 <epony> and laptops are in that category
12:03:00 <epony> SoC-ware and appliance consumer CPUs
12:05:00 <heat> mjg, unlikely to help anyway
12:05:00 <heat> you probably need to boot on a real raptor lake platform
12:05:00 <heat> instead of a 13th gen CPU and whatever devices QEMU/Amazon decide to make up
12:06:00 <mjg> heat: ec2 is handing out bare metal
12:06:00 <mjg> heat: or close enough to it
12:06:00 <heat> oh really?
12:07:00 <mjg> see 'metal' instances
12:07:00 <mjg> it's not all overprovisioned vms
12:07:00 <heat> that's really cool
12:07:00 <heat> so are they prohibitly expensive?
12:07:00 <heat> and how do they stop you from fucking shit up
12:07:00 <mjg> heat: i don't know pricing, i use them for free
12:07:00 <heat> :(
12:07:00 <mjg> ... provided i don't push it
12:08:00 <mjg> which i don't
12:08:00 <heat> is that a freebsd foundation benefit thing or what?
12:08:00 <epony> more like how hosting on facebook works, you get a free tier for trialware
12:09:00 <epony> gcp offers such too
12:09:00 <mjg> heat: there is a freebsd developer who for reaosns i don't specificaly know has a deal with amazon
12:10:00 <mjg> heat: gazilliard affiliate points or whatever it is
12:10:00 <heat> ah
12:10:00 <heat> cool
12:10:00 <epony> the cperciva is an aws s3 reseller
12:10:00 <mjg> that's the guy
12:10:00 <epony> the FreeBSD security "officer" who is a violin player
12:10:00 <heat> apple should gift something to freebsd devs too
12:10:00 <heat> like, extra icloud storage
12:10:00 <heat> does that sound good?
12:11:00 <epony> a box of pears and a bottle of rainbow wine
12:11:00 <mjg> :p
12:11:00 <epony> Apple is leasing cloud services from MSFT and AMZN
12:11:00 <epony> there is no iCloud, it's an application / leasing resale
12:12:00 <epony> IBM kicked them out
12:12:00 <epony> so Apple went out to ORACL and GOOG too
12:12:00 <epony> very versatilve, no own infrastructure
12:14:00 <mjg> heat: so if need be i can bench some onyx in a vm on otherwise unocuppied bare metal over there
12:14:00 <epony> you can do the same on OVH too for even lower costs
12:15:00 <mjg> i got the cost of 0
12:15:00 <mjg> is ovh going to pay me?
12:15:00 <epony> no
12:15:00 <epony> but it's not going to make others pay your costs
12:15:00 <mjg> then it's not lower cost for me
12:15:00 <epony> for everone else is
12:16:00 <micttyl> host your own cloud. put the servers in your friend's garage.
12:16:00 <epony> it's a decent competitor to aws
12:16:00 <epony> that wors even better
12:16:00 <epony> only the peering is garagehole
12:16:00 <epony> and you get cold drafts and leaks
12:19:00 <epony> also.. ovh has own international peering leases and decent backbone access, which lets them offer much lower costs than aws
12:19:00 <epony> instead of passing through 1 more tier like amzn does
12:19:00 <epony> and outside USA amzn does not own but rents
12:20:00 <epony> while goog owns data centres abroad
12:23:00 <epony> speaking of FreeBSD did Jordan Hubbard came back or did he manage to convince people in FreeBSD to become an Apple subsidiary and adopt the launchd systemd startup clones and the microkernels and insta-on features of locking up your server (you know Apple is not a server company)
12:24:00 <epony> and why don't people on the FreeBSD foundation staff not use FreeBSD but Apple and Windows laptops?
12:25:00 * epony shows the airbrake in the middle of the air
12:25:00 <micttyl> didn't hire such people i guess
12:26:00 <heat> mjg, that'd be cool but then if onyx doesnt boot i'll look super stupid :v
12:26:00 <epony> sounds to me, performance is not everything.. on the laptops ;-) must be some remote control from a peering corporate sponsorship
12:27:00 <heat> might as well become a BSD at that point
12:27:00 <micttyl> beside admin, it is somewhat serious problem since FreeBSD is advertised as also a good desktop OS unlike for example OpenBSD
12:27:00 <epony> that is inevitable
12:27:00 <heat> is OpenBSD advertised as a good anything
12:28:00 <micttyl> they don't officially appeal anything about desktop application
12:28:00 <epony> even Linux realised early on the ext to ext2 conversion that BSD is what works and has some wood knocking sanity reworks
12:28:00 <micttyl> FreeBSD however does so
12:28:00 <epony> FreeBSD has marketing
12:28:00 <epony> not as much as Linux but.. has
12:29:00 <mjg> heat: if it does not boot i simply wnt tell you
12:29:00 <mjg> heat: alternatively i'll mail the phoronix guy to write abouti t
12:29:00 <epony> Linux has become a foundation of foundations, forking all the time like rainshrooms
12:29:00 <mjg> heat: have not decided
12:29:00 <heat> lmao
12:29:00 <micttyl> two of my laptops runs only FreeBSD, no other efi boot
12:30:00 <epony> is phoronix not a newspam site?
12:30:00 <heat> mjg, have you seen the memes about turkish people saying turkey is the best all the way from berlin?
12:30:00 <epony> (of marketing and paid publications)
12:30:00 <heat> "freebsd is the best" - mjg, from RHEL
12:30:00 <mjg> heat: no
12:30:00 <mjg> i do genuinely htink a blogpost somewhere comparing perf of homebrew systems would be a fun light read
12:30:00 <epony> RHEL is a contender to FreeBSD, but FreeBSD has juice, and RHEL has money
12:31:00 <mjg> and one can be generous by not including openbsd on the list
12:31:00 <epony> RHEL without the IBM handholding is full of shit really, and the performance is mediocre as is
12:31:00 <heat> guarantee you I'll beat everyone
12:31:00 <epony> remember fsyncgate?
12:31:00 <mjg> probably true
12:32:00 <heat> the competition is either microkernels or understaffed monolithic kernels written by non-performance-nuts
12:32:00 <epony> https://wiki.postgresql.org/wiki/Fsync_Errors
12:32:00 <bslsk05> ​wiki.postgresql.org: Fsync Errors - PostgreSQL wiki
12:32:00 <mjg> i was considering writing about templeos vs performance, since terry made numerous mostly wrong claims
12:32:00 <mjg> but i expect people will give me shit for doing it
12:33:00 <sham1> "How dare you bully this mentally hurt man"
12:33:00 <epony> templeos has performance? that's like Wfw3.11 without the workgroups and without the windows
12:34:00 <heat> he dead
12:34:00 <mjg> should you poke around interwebz he reached "mainstream", and by that i mean there are people who are not programming and who know he wrote an OS
12:34:00 <epony> so.. holyc died too
12:34:00 <epony> plan9 has performance.. but fails on grid computing
12:34:00 <mjg> what i find problematic is the the dude is considered a /genius/ for writing one
12:34:00 <epony> and the performance is like 386BSD
12:35:00 <epony> that's a bit nonsensical (4chan idiocy)
12:36:00 <mjg> key for me is that he decided to not have user/kernel priv split and that was the intended end state
12:36:00 <mjg> for the end game
12:36:00 <epony> of the endspiel of the 16bit DOS epoch
12:36:00 <heat> we wasn't a "genius" but writing an OS is pretty hard
12:36:00 <mjg> "like commodore 64. like a motorbike -- you lean too much and you gonna crash"
12:37:00 <mjg> i think it would make for an interesting read why the mindset above is not fit for contemporary computing
12:37:00 <mjg> even for end user systems
12:37:00 <epony> like false analogies
12:37:00 <sham1> I mean, HolyC had some neat things about it
12:37:00 <mjg> heat: noq ustion he would be a good programmer if it was not for his mental health issues
12:38:00 <sham1> I mean, it's not like TempleOS was really meant to be "contemporary computing" in that sense. It was more a thing where it was meant as a "monument to God" in some sense
12:39:00 <mjg> his claims everyone who disageres is an MIT nigger suggest otherwise
12:39:00 <epony> it's not fit for contemproray systems, not because of his mindset, but because of yours not knowing about out-of-order and superscalar processors that came out to supercede and replace the previous models and also cancel the risc workstations with the same swift move (and to add some caching and what not else, like branch prediction and pipelining and new instructions sets)
12:39:00 <mjg> he explicitly claimed the cost of user<->kernel transition is too high to bare
12:39:00 <mjg> and linux is just dogshit for paying for it
12:39:00 <mjg> and so on
12:40:00 <mjg> the ring3/ring0 separation belongs in a mainframe!
12:40:00 <moon-child> I mean it does suck
12:40:00 <epony> it's like claiming DRI's CP/M and DR-DOS was revolutionary since it tried to do UNIX-like adaptations to the PCs before Intel had tested it out as "working"
12:40:00 <moon-child> but if you get rid of it you still have to provide a solution for security
12:40:00 <epony> while it's actually an adaptation of a system and nothing new
12:40:00 <mjg> moon-child: it is not free for sure, mitigations or not :)
12:41:00 <moon-child> but people obviously do fast stuff on linux. Esp. post-io_uring
12:41:00 <heat> well, you're complaining about the mental behavior of a mentally ill person
12:41:00 <mjg> i'm not complainig about terry
12:41:00 <heat> good "programmer" sure
12:41:00 <epony> it's ring -3 and there is also -4 now
12:41:00 <mjg> i'm comlpaining about people tauting him a genius
12:41:00 <heat> we wrote a full OS and toolchain and shit
12:42:00 <heat> s/we/he/
12:42:00 <heat> inb4 im terry
12:42:00 <sham1> "Je Suis Terry"
12:42:00 <mjg> perhaps i could try to clarify this way
12:43:00 <mjg> non-tech problem: people love to claim FUCKING GENIUS when someone does something they have no clue about
12:44:00 <mjg> tech problem: intentional lack of priv separation in templeos would make for an informative post why it is not fit for sensible use today for end user devices
12:45:00 <mjg> trivial examlpe would be what happens when a program crashes
12:45:00 <mjg> do you want to need to reboot?
12:45:00 <mjg> and then another one about security
12:45:00 <mjg> even your fucking phone uses a lot of privsep
12:45:00 <heat> oh yeh totally
12:46:00 <heat> doesn't a phone have like 3 separate priv layers + secure world or whatever
12:46:00 <mjg> that said, i think there is sensible & educational things to say here, while not shitting on terry as a parson
12:46:00 <epony> here is the ring -4 https://en.wikipedia.org/wiki/One_ring_to_rule_them_all?useskin=vector
12:47:00 <epony> and they call the service personnel in the Apple shops "geniuses" too, but that's not really a genius either
12:48:00 <mjg> that is a very different subject tho, innit
12:48:00 <epony> your phone uses shit separation where your baseband modem is a SoC on its own and has memory access to the embedded CPU SoC where it's attached
12:48:00 <epony> so the separation in the phone is.. a complex and mostly laughable matter
12:49:00 <heat> tbf he also had a point
12:49:00 <heat> you want to get around the kernel<->user barrier as much as possible
12:50:00 <heat> such that io_uring and eBPF everything are a thing
12:50:00 <heat> you can literally inject """safe""" code into the kernel
12:50:00 <epony> https://en.wikipedia.org/wiki/Baseband_processor?useskin=vector#Security_concerns
12:51:00 <epony> https://en.wikipedia.org/wiki/DMA_attack?useskin=vector
12:54:00 <epony> that's what the browser is aiming for with the _Rust_ implants in the kernel too
12:55:00 <epony> don't think it's for making your kernel run faster or safer, it's for tapping the browser with kernel support bypassing system premises / module support for hypervisors etc
12:56:00 <froggey> mjg: as someone who's written a something similar to templeos... yeah you don't have to be a genius for that and it's kinda weird that people elevate him to that kind of level. dedication and lots of free time are the important factors
12:56:00 <froggey> I'd also be interested in a performance comparison of hobby systems, but I guess that might get difficult for some of the more out-there systems
12:57:00 <epony> also there is not a lot of prophecy in the "holy" part of it ;-) it's more like inability to understand the metaphors of religion in everyday life
12:57:00 <mjg> froggey: i see the 'genius' thing everywhere
12:57:00 <mjg> froggey: not just about terry or programmming
12:58:00 <mjg> it's pet peeve of mine
12:58:00 <mjg> anyhow, i'm afraid for least unfair perf comparison one may need to stick to unix clones
12:58:00 <epony> and a lot of infidels too
12:59:00 <epony> UNIX has pefrormance, it's WIN/DOS that is blocking on floppy access and printer spool
12:59:00 <mjg> the good news is that this is what mosto f the systems seem to be
13:00:00 <epony> UNIX was designed to work with 10-100 terminal lines (even 256)
13:00:00 <epony> Windows was designed to run with one idiot behind it..
13:03:00 <epony> it's not much of a performance if you're running at max CPU speed a 1 application in a https://en.wikipedia.org/wiki/Unikernel?useskin=vector it's about as single address space as can get
13:04:00 <epony> here is your best "speed" demons, most of them defunct and abandonware https://en.wikipedia.org/wiki/Single_address_space_operating_system?useskin=vector
13:12:00 <sham1> WinDOS also blocks me from writing to a file that's opened elsewhere
13:12:00 <sham1> Literally unusable
13:13:00 <epony> "Windows has performed and illegal context switch and needs to switch back to safe mode, please wait while we reboot your autopilot" --the macAir port
13:13:00 <epony> and and and
13:14:00 <sham1> And yeah, it actually is annoying that it forbids me from saving while documents are open. Makes synctex workflows annoying, for example
13:14:00 <epony> https://en.wikipedia.org/wiki/Scheduling_(computing)?useskin=vector#Windows
13:14:00 <epony> half-duped simplex networking for the masses and their molasses
13:19:00 <kaichiuchi> hi
13:19:00 <kaichiuchi> froggey: i think it's the fact that he was homeless and a schizophrenic and was still able to pull it off
13:20:00 <froggey> I'm pretty sure he lived with his parents for most of the development
13:20:00 <epony> the glowies stuff was not nice
13:20:00 <epony> and being homeless 5 minutes before the train resolves the problem is not really homelessness
13:20:00 <kaichiuchi> i dunno
13:21:00 <epony> alcohol and narcotics have spoiled lots of promising youths
13:22:00 <kaichiuchi> mjg: you don't need my advice, but I wouldn't advise a blog post or something about him
13:22:00 <kaichiuchi> because you're just going to get shit on massively
13:22:00 <kaichiuchi> sometimes, to hold a pen is to be at war
13:23:00 <epony> and to hold an emulator is to be at universe creation, not really but.. Elon believes the same shit
13:23:00 <kaichiuchi> and re "genius"; when I was growing up, computer laymen (my parents, my elementary school teachers etc.) kept calling me a genius because I was able to use a computer, and 'fix' computers
13:23:00 <kaichiuchi> big mistake.
13:24:00 <froggey> the internet loves its darlings
13:24:00 <kaichiuchi> I'm not saying terry wasn't smart
13:24:00 <epony> face it, you could have become a genius if you had read a set of textbooks (or two) about computers
13:25:00 <kaichiuchi> but I think people look to him in the sense that he was severely mentally ill and still pulled off something fairly complicated
13:25:00 <epony> but it's not consumer tech-pulp books that make it happen
13:25:00 <epony> and if you believe that people like Carmack are geniuses..
13:26:00 <kaichiuchi> it's meant to be an inspiration
13:26:00 <epony> that's not severe mental illness, though
13:26:00 <mjg> kaichiuchi: 13:32 < mjg> but i expect people will give me shit for doing it
13:26:00 <kaichiuchi> oh
13:26:00 <kaichiuchi> sorry
13:27:00 <kaichiuchi> heh
13:27:00 <mjg> maybe i'll write a post how i heard heat is a genius for writing onyx
13:28:00 <kaichiuchi> i'm just saying, I'm not a schizophrenic, but try doing anything productive whatsoever when you're on dopamine antagonists, and even if you aren't, hearing or believing that the CIA is going to come kill you and your whole family
13:28:00 <epony> well, if you start with pills and vodka now, in a couple of years that might change ;-)
13:28:00 <kaichiuchi> antipsychotics render your only thought for the day "should I have eggs or french toast for breakfast" before sleeping 18 hours
13:29:00 <epony> the CIA is capable of doing that
13:29:00 <epony> and has been doing THAT to some people (and their families)
13:30:00 <epony> maybe not to goofs and drools but still
13:30:00 <epony> the holy-dead-c scrolls is the real mental disorder
13:31:00 <epony> and it's probably been minor in his case with dellusions and cyclophrenic patterns, like a high-functioning autism of some sort
13:31:00 <epony> the bad thing is peole don't have access to quality medication and treatment in the expensive capitalist health"care" system
13:33:00 <epony> so it's mostly a sadness to his family and people in such position, not a really big loss to applied sciences
13:34:00 <epony> so might as well stop believing the Allan Kay "kerpow" talks about education as a means to sell overpriced computers
13:47:00 <Ermine> mjg: re templeos: isn't user supposed to pray if some program crashes?
13:48:00 <micttyl> it was an interesting discussion, so i looked around and found that there is an opinion about donation money goes to somewhere else: https://web.archive.org/web/20210318184422/https://lists.freebsd.org/pipermail/freebsd-chat/2018-March/007181.html
13:48:00 <bslsk05> ​web.archive.org: FreeBSD's politics problem
13:49:00 <epony> there is a really scary part to this though, not in the software, but in the hardware, or rather.. the wetware https://en.wikipedia.org/wiki/Lobotomy?useskin=vector#Prevalence
13:50:00 <mjg> micttyl: while i don't know the full accounting of freebs foudnation financials
13:50:00 <mjg> micttyl: i can tell you i know for a fact that there is money going towards development of the system and various hardware
13:51:00 <epony> 'freebs' sounds very funny
13:52:00 <mjg> micttyl: on that point, a dev machine for i'm using for freebsd scalability work got extra 280G of ram 2 weeks ago to enable some benches to be ran
13:52:00 <epony> the management of funds is obviously going well since the FreeBSD system is competitive and significant as a contender with Linux's popularity, although with a much more technical quality to it in the 1995-2010 period
13:53:00 <mjg> interested parties can find sponsored work in 'sponsord by' tags in the commit log
13:53:00 <mjg> s/in/by/
13:53:00 <epony> that sucks
13:53:00 <micttyl> i am not used to this sarcasm
13:54:00 <epony> sponsored commits and log messages smells a lot like commercialised development
13:54:00 <gog> brought to you by Carls, Jr.
13:54:00 <epony> Linux has been eating shit about the corporate control for the last 25 years
13:55:00 <epony> and the useds starting eating the same with changes in the interests of automation and data centre provisioning and binary interfaces and such for a long time too
13:56:00 <epony> and you can say GNU uses the same embrace extend and extinguish practices that Microsoft does, only not for binary only distribution..
13:56:00 <micttyl> given the reality some people need to make living, i don't mind of corporate interference. i think the argument should start from: "how is this control bad"
13:57:00 <epony> you're missing the point of free software though
13:57:00 <epony> that's shareware and source available trialware
13:57:00 <gog> brought to you by carls, jr.
13:57:00 <micttyl> epony: you're right indeed
13:57:00 <epony> and the points in the BSD have been non-commercial UNIX
13:57:00 <mjg> Ermine: i don't know anything about that
13:58:00 <mjg> Ermine: i only remember explicit claim that crashes happen and you have to deal with it
13:58:00 <mjg> and that address space and priv separation is too costly to do
13:58:00 <epony> I think GNU should keep a more competitive stance in the interest of the users and the public, rather than its own distribution "rights" and "protectorate" as inheritance to its foundation chairs and staff
13:59:00 <mjg> all stated prior to meltdown et all mind you
13:59:00 <mjg> et al
13:59:00 <epony> and FreeBSD could benefit from less commercialisation and more "due diligence" into developers being UNIX and BSD people instead of Apple and Windows affiliates
14:00:00 <epony> but these points are minor
14:00:00 <gog> cache timing attacks are irrelevant without priv separation anyway
14:01:00 <epony> what's bad is the dependence on Linux import in the BSD world to keep support for graphics and heavy compilers and monster dependencies and reduction of quality in applications on accelerated pace that target only Linux and that being a second target after Windows (browsers and IDEs being the prime abuse of system resources by bypassing the system premises and doing double overhead of memory management in application space)
14:02:00 <Ermine> gog: may I pet you
14:03:00 <micttyl> we have linuxkpi to borrow linux features
14:04:00 <epony> and the other really shitty thing is using software middleware modeled and designed to target commercial and single vendor CPU designs that are not at all the mainstream and business ready office and server class machines, but proprietary smaller and less capable ones, resulting in penalty in performance on most in use computers for really no good reason
14:07:00 <micttyl> oh like in-house muh-way standard?
14:16:00 <gog> Ermine: yes
14:19:00 <kaichiuchi> gog: may I have some chili
14:22:00 <gog> kaichiuchi: yes
14:22:00 * kaichiuchi eats chili
14:22:00 <epony> here is a generic details listing of the performance and efficiency cores premises which you will find on Intel's latest laptops too
14:22:00 <epony> https://en.wikipedia.org/wiki/Global_task_scheduling?useskin=vector#The_problem_that_big.LITTLE_solves
14:23:00 <epony> so it's expected that systems that don't make use of that will "suck" for some time as efficiency
14:24:00 <epony> so might as well wait it out or program your way into it, or change form factors / CPUs or OS for the initial adoption "mandate"
14:24:00 <epony> in short: laptops are becoming frisbees as tablets and phones
14:29:00 <epony> and that explains the LLVM/MLIR and JVM/BC/IR/CLR controlled runtime middleware too, that optimises for that model of programming
14:30:00 <epony> the main goal: application stores and sucbscriptions, at the cost of a long transition period, application rewrites and reimplementations, and higher prices, and lower performance.. but better battery life, and more control of business over the middleware
14:31:00 <epony> what's not to like about "commercialisation" and sponsorship with "command and control" strings attached to it
14:32:00 <epony> the Chinese new year reminds the "may you live in interesting times" fortune cookie (curse)
14:32:00 <epony> it's going to be much fun
14:33:00 <epony> eventually, OSes will improve, and applications too
14:33:00 <epony> as for the battery life on servers.. not so sure ;-)
14:35:00 <mrvn> moon-child: God fearing users don't do bad things. TempleOS needs no security.
14:36:00 <epony> that's exactly what Lisp has been saying for 60 years
14:36:00 <epony> the Interweb does not believe list processor tears
14:37:00 <mrvn> heat: phones have so much priv. separation and security that you don't even bother. You just inspect the encrypted data and reconstruct the voice from correlations.
14:38:00 <mrvn> It's too bad the security is flawed.
14:39:00 <epony> it's due to deliberate choices in hardware and middleware
14:39:00 <epony> and various communications protocols
14:39:00 <epony> phones are not securable without control of the built-in baseband modem
14:39:00 <epony> it's a botnet
14:40:00 <epony> like browsers
14:42:00 <epony> also the OS does not control much of what the other parts of the SoC are doing and it has direct access to the data lines in CPU space
14:45:00 <epony> not a surprise to have https://en.wikipedia.org/wiki/WARRIOR_PRIDE?useskin=vector and https://en.wikipedia.org/wiki/Mobile_malware?useskin=vector#Notable_mobile_malicious_programs and https://en.wikipedia.org/wiki/Vault_7?useskin=vector#Apple_products and https://en.wikipedia.org/wiki/Signalling_System_No._7?useskin=vector#Protocol_security_vulnerabilities
14:45:00 <epony> it's a lot of lies and marketing BS
14:48:00 <mrvn> micttyl: when a project starts paying people there is always a risk those people will then work for getting payed more instead of working for the project.
14:51:00 <micttyl> mrvn: that a big pitfall
14:51:00 <mrvn> Soon you are paying someone to handle paying people.
14:55:00 <micttyl> mrvn: do you worry about your low-paying employee moving to another project?
14:55:00 <micttyl> mrvn: do you want to stop them?
14:56:00 * Ermine pets gog
14:56:00 * gog prrr
15:12:00 <heat> mjg, i want to be a misunderstood genius after I pass away relatively young
15:35:00 <sbalmos> heat: your cause of death will be misunderstood. close enough?
15:51:00 <zid`> sbalmos: I'll just write it down as "misadventure", saves the family from learning he was a pervert.
15:56:00 <heat> misadventure in thailand
16:16:00 <mjg> anybody knows a good place where one can ask cpu engineers about something
16:17:00 <mjg> now that i asked the above, perhaps i can try intel forums
16:21:00 <heat> lmao
16:23:00 <mrvn> micttyl: I want to have enough people to take over every job so that that isn't a worry.
16:24:00 <mrvn> micttyl: having firms pay developers directly because the firm uses the project internally or as part of their busyness is also great. Better than donations to pay someone.
16:47:00 <mrvn> heat: die young and leave a beautiful corpse?
16:49:00 <mrvn> mjg: or just ask here
16:50:00 <mrvn> there are a few cpu builders in here although that might not be what you need
17:00:00 <kof123> he said he wants to be a misunderstood genius *after* he passes away. this is afterlife only, he'll still be the heat you all know and love in the meantime.
18:26:00 <Bitweasil> mjg, depending on what it is, asking here is probably the best source I know of, short of actually knowing people at Intel or AMD or such.
18:56:00 <heat> mjg is beside himself looking for the 13th gen's verilog as a way to find out why freebsd doesn't boot
18:57:00 <mjg> genz humor calmed me down
18:57:00 <zid`> I want an airfix 8086
18:57:00 <zid`> cut all the transistors out, glue them in
19:02:00 <sham1> Now I want a gen z processor
19:03:00 <zid`> I can get you a z-machine
19:03:00 <zid`> is that good enough
19:03:00 <epony> you get an Rpi3
19:03:00 <epony> that's gen Z
19:04:00 <heat> does the z in z-machine stand for zid
19:04:00 <zid`> no
19:04:00 <zid`> zork
19:06:00 <sham1> Zoinks
19:06:00 <zid`> no that's a mystery machine
19:06:00 <sham1> That'd be rather myserious
19:06:00 <epony> the Z in IBM Z stands for Ze IBM
19:07:00 <mrvn> does it have a dog?
19:07:00 <epony> but since it's a CISC, you don't like it
19:07:00 <kof123> they had soem kind of eye bee and i forget the M screensaver
19:07:00 <epony> for you only RISC works, the CISC is broken for you
19:08:00 <sham1> That's just the defining feature of CISC, not working
19:08:00 <sham1> And being inelegant
19:08:00 <zid`> but being fast and easy on the eyes
19:08:00 <zid`> like a racing snail
19:09:00 <epony> that's what the RpiZero is
19:09:00 <epony> a zero performance best-in-show znail
19:10:00 <epony> with Apple technology inside
19:34:00 <epony> btw, you don't want FreeBSD really, for GenZ you want Redoz and micro-zernels
19:34:00 <epony> and it's not gen13, it's genZirteen
19:36:00 <demindiro> I finally got the design of my cache right
19:36:00 <demindiro> Probably
19:36:00 <demindiro> 6525 insertions(+), 3640 deletions(-)
19:36:00 <demindiro> Only a mere 17628 loc total
19:37:00 <epony> it's supposed to be approx 0 lines and be in hardware
19:37:00 <demindiro> I'd rather shoot myself than try implement my cache in hardware
19:37:00 <netbsduser> what sort of cache?
19:38:00 <demindiro> Cache for my custom filesystem
19:38:00 <epony> zCache
19:38:00 <demindiro> Which was supposed to be pretty simple but alas
19:38:00 <mrvn> distributed caching?
19:38:00 <demindiro> cache invalidation + redirect-on-write + compression + whatever with concurrency turns out to be hard
19:38:00 <epony> you've been tainted by zFS on the future to be announced version z23 of zOS
19:39:00 <epony> especially if you're writing it in Lizp
19:40:00 <Ermine> where genz humour?
19:40:00 <demindiro> What humour?
19:40:00 <epony> when it's ported from Lizp to z/LIZP
19:40:00 <epony> remember, no slashing no fun
19:42:00 <epony> which lz4 did you use for comprezzion?
19:43:00 <mrvn> I just started building a fuse FS that uses a 1TB tmpfs as cache for a rotating disk array.
19:44:00 <demindiro> 1TB seems a bit excessive
19:44:00 <mrvn> or maybe just memory in the fuse process
19:44:00 <epony> why are you rotating the array? ;-)
19:45:00 <epony> you don't have that much RAM
19:45:00 <epony> especially on on Rpi5
19:45:00 <mrvn> why?
19:46:00 <epony> is that supposed to be some version of the arc cache? https://en.wikipedia.org/wiki/Adaptive_replacement_cache?useskin=vector
21:49:00 <\Test_User> "Assumption of CODE and DATA occupying the same space, (at least with PAE NX bits and Paging units not used) allows ROGUE/virus like code in the first place to take advantage of the machine. INTEL Specs even say this. " (near the end of https://wiki.osdev.org/Segmentation)
21:49:00 <\Test_User> why exactly is this? just the typical "rwe memory is dangerous bc program could be tricked into executing arbitrary code if it's got the right bugs" stuff or some actual cpu issue allowing more than what would normally be allowed
21:49:00 <bslsk05> ​wiki.osdev.org: Segmentation - OSDev Wiki
21:50:00 <heat> sounds like bullshit
21:50:00 <heat> at least very non-wikipedia-like
21:50:00 <heat> but yes rwx bad
21:51:00 <heat> "Microsoft still is plagued with this problem,despite having NX bits enabled even in the latest OSes. " bs?
21:52:00 <\Test_User> yeah that doesn't quite add up bc it says "with PAE NX bits and Paging units not used" but then that has them