MINTIA (not vibecoded)

[twilit smelt]

its wikipedia its not a billboard

[twilit smelt]

thats not that big

i could lift that

small computer really

(DEC Alpha mainframe from 1992)

[shadow ridge]

looks like a fridge

[balmy zealot]

I see, it turns out the boundaries are softer than I thought. Thanks.

[twilit smelt]

i need to make a logo for XRSI or XR/corp or whatever its called

ive wanted to do some aesthetic work for a while an d i think today may be the day

[twilit smelt]

i want the firmware to draw a cool sidebar containing various system info which OSes avoid drawing over until their DE is ready to come up

mock up

much more stuff will be in that sidebar like (complete list):

• XRSI logo
• representation of the system its running on (thats the only thing there in the mockup)
• rectangles for each of the NUMA nodes which say "NODE 0: 4096KB" or however much memory
• NUMA node rectangles also contain little icons for each of the CPUs which start out saying "IDLE" until kicked by the system software at which point itll say "ACTIVE"

[balmy zealot]

Does your hardware platform support NUMA?

[twilit smelt]

yeah

i dont simulate the latency yet though

[balmy zealot]

I'm curious to know whether you break the nodes down into subnodes or not

[twilit smelt]

i do not

its up to 4 nodes with up to 4 processors each

[twilit smelt]

on the software side itd probably be a good idea to stop conflating kernel nodes with hardware nodes

having a different number of virtual nodes than physical nodes could be desirable for scalability

[twilit smelt]

working on it

[twilit smelt]

hmm maybe the sidebar should be on the left

[shadow ridge]

make it float around like a dvd screensaver

[twilit smelt]

left is WAY better

if you have a portrait display it puts the sidebar at the bottom

ignore the text being weird thats just placeholder text until theres a bitmap to draw there

hmm itd be cooler at the top tho

top looks stupid too

so i just made it always be on the left

[twilit smelt]

drawing run length encoded xrstation image

the rom is only 128KB so this is actually quite difficult to fit

luckily it seems it would actually look better if that bitmap were a bit smaller

so thats an easy way to save space

[warm mural]

dunno if that might help but there are simple compressed image file formats

like qoi

dunno how much space that will save you though

[twilit smelt]

it says its usually worse than png and the png of that is already 25k

my RLE is 8k

sooo

thas not surpsoed to hapen

[twilit smelt]

@icy bridge when im done with this it would b sick if you modified your linux port so that the video terminal stays out of the way of the firmware's sidebar, the idea is that it stays visible until the OSes desktop environment comes up, so like a boot log or early terminal would ideally not obscure it

i think it would look really cool

when its done itll like display the name of the booted OS, the status of all the processors, etc

and a bitmap image of the computer its running on

[twilit smelt]

as here

[shadow ridge]

cant you make a vectorized version of that and render that instead

[icy bridge]

i think the way you're supposed to do that kinda thing with linux is using a non-verbose boot (which makes it not touch the framebuffer at all)

[warm mural]

Looks cooler bitmapped

[icy bridge]

but I'll look into it

[shadow ridge]

but it's eating lots of memory

[twilit smelt]

theres gonna be some new entries in the device db struct for the width and location of the sidebar so you know what to stay out of the way of

[twilit smelt]

only a few kb

i have a good rle

about to get better i just had a good idea

[shadow ridge]

i bet you could encode it as vectors and have it take like 200 bytes at most

[twilit smelt]

it was 8k now its 5k and with this next change i wonder if i can get it to 3k

also i realized the fonts encode a bunch of glyphs for ascii codes >=128 which i dont need and can trim off

[icy bridge]

sounds good, I think the best way to do that would be to communicate a display rectangle

[dense vigil]

how can rle be good or bad

also its like mostly a black rectangle so ofc rle wins here

[twilit smelt]

ud be surprised how many bit tricks you can play to get it tinier and tinier

its fun actually

[dense vigil]

do u compress the color somehow?

u can probably use a 1 bit palette lol

[twilit smelt]

its more about reducing the metadata but im going to do that too

[twilit smelt]

no i cant

that would look rly bad

[shadow ridge]

2 bits

[twilit smelt]

the antialiased pixels r important

theyre all grey

here

[dense vigil]

ah ok

[twilit smelt]

so like i could do a 16 bit palette of greys

or something

and in fact im going to look into making each rle image encode a palette beforehand and then use indices into the palette

[shadow ridge]

i see like at most 8 colors there

cant u do antialiasing afterwards

[short owl]

They're saying they might want more color for later

[shadow ridge]

then they can handle that differently

[short owl]

I'd say 16 colors are more than enough to make this kind of thing

[shadow ridge]

hmm

what if u used like 4 colors and dithered it afterwards

[twilit smelt]

overcomplicated

[shadow ridge]

how is dithering more complicated than RLE

actually if you dont do RLE and use 2 bit colors you still get like 16 kib per 256x256 image

hmm

[short owl]

I got a crazy overcomplicated idea for compressing this kind of image

But i better shut up

[twilit smelt]

no say it

[shadow ridge]

dont you dare say it

[twilit smelt]

ignor him

[shadow ridge]

ignore this guy he doesnt know what he's talking about

he's happy with his RLE that can only do like 5 kib

let him live in blissful ignorance

[short owl]

Okay but hold up i'm doing something rn

[twilit smelt]

my current idea im doing is runs of runs

metaruns

[delicate dome]

I don't recall what method this was again but use the decompressed data, a pointer within it and a length, to get both RLE and pattern compression

[short owl]

So since this image is mostly just big areas of one single color you could fierst break it into a grid of square cells of k size on each side and start merging cells that don't have any sub-cell details (so they're just one solid block of some color) into rectangular or square cells and on cells that do have sub-cell details you could do normal rle

(you can do the merging in blth dimensions)

[warm mural]

lz4 it

Not that hard to implement and very good

[short owl]

Then you just describe "heres a big block of some color that goes from cell X1,Y1 to cell X2,Y2" and the ones that couldn't be merged this way can be RLE'd and also have a small metadata saying the location on the grid of these cells

This will probably save up a lot on images that aren't too noisy overall

[short owl]

@twilit smelt so what do you think

[twilit smelt]

thats an interesting buge

[twilit smelt]

it sounds very fancy and good

but ur right its overcomplicated lol

50 good boy points to whoever can decode this format without me giving them any hints

decodes into this

[twilit smelt]

kind of a cool looking bug

waoe

woohoo

down from >8k with the original rle

[warm mural]

still too big...

[twilit smelt]

check THAT out

i got it down to 2k by encoding all the colors as 4 bit indices into a palette

and all of them are packed at the start

the run control bytes come after

format:

- offset to run control bytes from start (2 bytes)
- 4-bit palette (16 bytes)
- nibble pool (N bytes, packed 4-bit nibbles)
- run control bytes (M bytes)

[mortal thunder]

cant you configure the linux kernel console framebuffer driver to start drawing offset from an x coordinate?

[twilit smelt]

thats only 500 bytes worse than DEFLATE gives me btw

[mortal thunder]

like framebuffer->start += x * sizeof_pixel; framebuffer->width -= x;

[mortal thunder]

what if you introduce long and short runs

bytes of the format xy where x is between 1 and 15 (pixels) of palette index y

and 0y xx where x is the length + 15 and y is the palette index

[twilit smelt]

my runs look like this

--   7   6   5                             0
-- +----------------------------------------+
-- | L | G |             LENGTH             |
-- +----------------------------------------+
-- |                LENGTH                  |
-- +----------------------------------------+

where the first byte contains the low 6 bits of the length biased by -1 (so a zero means length of 1), G=1 means the run is a run of one nibble from the pool, G=0 means the run is a run of consecutive literal nibbles from the pool (this is used when theres a failure to compress due to a long series of un-alike pixels), and L=1 means the length is greater than 64 and therefore the second byte is needed

if L=0 then the second byte doesnt appear and the run is encoded in a single byte

the lua script i wrote for encoding this is 200 lines

and is extremely ugly

ive heard of using a bitwise RLE for encoding bitmap fonts and in fact this is how fonts were encoded in the original macintosh's rom for example

im not gonna bother with that lol

bitwise RLE is really easy though actually

@warm mural @mortal thunder do either of you know what the easy way to do bitwise RLE is

without googling

if you google youre lame

[mortal thunder]

i know of a way to encode continuous runs of 1 and 0 bits really compactly

which is that you specify the length of the next run, start at 0 and switch between 0 and 1

so like 00 05 06 07 decodes into 111110000001111111

(i swear i havent googled it, i just kinda remembered something associated with how the original GB pokemon games store their character sprites)

[twilit smelt]

yeah inversion points

[mortal thunder]

bitwise rle seems pointless unless your fonts are really big?

like if your fonts are 5x8 for example there's little need at all

you can get the same storage capacity in 5 bytes

[rare epoch]

you used 4 bytes to encode 18 bits though 😭

[mortal thunder]

without even needing to compress them

[mortal thunder]

thats not the point

i wasnt gonna type 240 ones and 255 zeroes to demonstrate it better

[rare epoch]

are those cases even common though? I’d have thought normal data wouldn’t have such regularity, at least color data
what would it be used for?

[twilit smelt]

i think apple's encoding was more like this

  7  6        4  3  2        0
+---+----------+---+----------+
| S |  Length  | S |  Length  |
+---+----------+---+----------+

where it was packed so that the run lengths were 3 bits (1-8) and the S bit decides whether you actually invert or not

[mortal thunder]

this might work better for smaller fonts

than what i thought

[twilit smelt]

so your example would be encoded as

1 100 1 101 1 110

which is 12 bits to encode 18 bits, a win

[short owl]

It's somewhat inspired by video codecs lol

[warm mural]

i havent looked much into compression but i guess one way would be you have one bit for the data and 7 bits for the counter/length

iprogramincpp's way seems better

[twilit smelt]

visual console continuity into mintia2

dont be fooled the firmware is DEAD! by then and gone

none of it is left

this is a visual trick to be pretty

fireworks

made it so the crash screen 'shutter' effect doesnt shutter the firmware side box

the fonts dont need to match up for the console continuity to work because the cursor is reported as pixel coordinates rather than character coordinates, so it is font agnostic

theres mintia2 using sun gallant demi while the firmware console was using haiku console font

[rare epoch]

it looks really cool

idk what else to say

I like how much larping you do (in a literal sense, not in a derogatory way), it gives the project a lot of character

[ocean valley]

quite cool indeed :3

[warm mural]

Did that system image on the side come from a specific firmware or did you invent that

[twilit smelt]

wdym ssystem image on the side

[warm mural]

Like having the picture of the machine

[twilit smelt]

i made that up

[warm mural]

In the firmware

[twilit smelt]

i made the image with google sketchup

plus a bunch of scripts to RLE encode it

[warm mural]

Yeah but the concept didn't already exist right

[twilit smelt]

it didnt

ive never seen it at least

[warm mural]

yeah ok

[twilit smelt]

its the same image used on the cover of the handbook

[warm mural]

It looks cool

[warm mural]

Gives me next vibes

The firmware

[rare epoch]

My only criticism is that what’s the point of it? Like it takes away usable space just to be able to display a cool image

[twilit smelt]

there will be more info in the sidebar

[rare epoch]

oh I see

[twilit smelt]

and also it will go away when the DE is up

itll only be there during boot logs and like text mode use of the OS

[warm mural]

Are you gonna bring back the mintia console eventually

[twilit smelt]

yeah its only gone bc i dont have an IO system yet lol

itll also position itself in the middle of the rectangular area to the side of the sidebar to avoid obscuring it

[rare epoch]

when is my kernel gonna be cool like that

i feel like there is like several months before I can start running DOS apps on top of it

[warm mural]

Don't think that's gonna look good

Too much clutter maybe

[twilit smelt]

it will look good

its not necesarily gonna look exactly like the old mintia console did

[cursive charm]

is your OS gonna be like OS/2?

[rare epoch]

more like Windows 3.1 / 95

it is a DOS based OS

maybe i should’ve made an OS/2 clone instead

given how it was meant to be a “better Windows than Windows”

I don’t know anything about OS/2 design or internals though

[cursive charm]

i haven't looked into OS/2 or win9x much

[rare epoch]

like one of the design requirements of my OS that everything uses DOS

drivers, IO etc all done through DOS and BIOS

which I know is bad design

because even Windows 3.1 realized this was a bad idea and shipped 32-bit disk drivers

but I want to rely on DOS as much as possible

[cursive charm]

i thought that was a windows for workgroups 3.11 thing

[rare epoch]

I don’t think so

I honestly think my idea is so cool

I hope the execution ends up being well

[warm pine]

there is a book OS/2 internals

it is quite unixlike in architecture

[twilit smelt]

dont think thats normal

[mortal thunder]

why are the fireworks so small

theyre like boxed in

[mortal thunder]

people have been trying to get away from DOS, why the hell would you want to rely on DOS

[twilit smelt]

bc i made them smaller

long time ago

wanted to be able to see logs and stuff while theyre going

there are no fewer particles the view port is just smaller

[warm mural]

poor fireworks

they're trapped...

[rare epoch]

because I love DOS and I wanted to make something that uses it

and with memory protection it’s not that bad

like the biggest issue is that a lot of IO operations are going to be slow

[cursive charm]

i think i've once heard of someone who really loves win9x and doesn't like NT, lol

at least this part of their winME video sort of indicates that, i guess:

yes, apparently, a "poorly written driver" can easily bring down NT, but not win9x somehow, according to this framing

[warm mural]

I really doubt that lol

[cursive charm]

i could link the whole video, but even in context, this just felt really odd to hear the first time i watched it

i guess they were sick of people claiming NT4/2000 was infinitely more reliable than contemporary versions of 9x?

[rare epoch]

honestly the more I learn about how 9x the more I appreciate it

I don’t think it’s as much of a hack as people claim / think, it’s a proper 32-bit multithreaded operating system that rarely uses DOS for anything and nicely extends it

it has some really major protection issues like exposing IDT but those are due to silly backwards compatibility reasons

[cursive charm]

i'm curious to see how its dynamic linker works

supposedly, its implementation of ExitProcess() didn't terminate all but the calling thread prior to performing process shutdown as opposed to NT

that, or raymond just said it worked different from NT somehow

[rare epoch]

And I find it heavily unfair that Windows 3.1 is just treated as an interface that runs on top of DOS

It’s literally a full 32-bit preemptive multithreading kernel with memory management and stuff

(in enhanced mode)

[warm pine]

by any measure windows 9x is a wonderful design given the extreme path-dependence it had to contend with

[rare epoch]

I love 9x

[cursive charm]

i don't think it's even true for, like, windows 1.0

[rare epoch]

yeah, but it’s way less true for something like 3.1 enhanced mode

[rare epoch]

my design diverges from 9x in that my kernel uses DOS for IO and drivers

instead of shipping its own drivers

but other than that I’m trying to be similar to 9x design

[warm pine]

can we see it? i'm interested how this ends up looking

[rare epoch]

currently there is nothing you can see - no interface and no apps for it, I’m focusing on the basics of the kernel before those
there isn’t even a crash screen or kernel debug output

it isn’t anywhere near being ready

but I had posted an old source code dump here:

#lounge-0 message

I tested it from kernel space (and user space) and I could use DOS/BIOS for IO and drivers, so the basic idea works

I’m now rewriting major parts of it to be more robust and support running DOS apps on my OS

if you want to compile it you need to use OpenWatcom in DOS, run “windy” to enter the kernel (just like how you run win to enter Windows 3.1)

Anyway I shouldn’t be talking about this here, if you want to continue you can comment on my thread - it’s also here, search for the name Windy

I feel like I hijacked @twilit smelt’s thread which is not nice

to hyenasky - I’m sorry for hijacking your thread to talk about my project, won’t do that again

[cursive charm]

is it #1495511907472248863?

[rare epoch]

yes

[mortal thunder]

I'm pretty sure POSIX doesn't require that to happen

Like, destructors get called even while other threads are still running and stuff

[cursive charm]

i was talking about win9x, not POSIX

win32 isn't POSIX, so i don't get how that's relevant

[mortal thunder]

I'm just saying that the "other" design also doesn't handle the case where you have other threads running and you call exit

[cursive charm]

well, i only brought it up because NT doing it was a point of criticism from a project on github that i was researching

[mortal thunder]

https://devblogs.microsoft.com/oldnewthing/20070503-00/?p=27003 i think this is the article you're taking about

[cursive charm]

yeah

[rare epoch]

I love Raymond Chen articles

[cursive charm]

for the record, the project in question is, of course, the operating system design review by eillot killick, and specifically under "Process Meltdown," they criticized the way ExitProcess() dealt with outstanding threads on NT

apparently, this was the only part of the document i could find regarding how the "other" design handles process shutdown:

The GNU loader doesn't implement any such startup performance hack to forgo locking on process startup. The absence of any such mechanism by the GNU loader enables threads to start and exit at process startup or within a module initializer. The same is true for process exit. Therefore, the GNU loader is more flexible in this respect.

this is from the end of "Windows Loader Initialization Locking Requirements"

simply put, there's too much from the document alone i could try to pick apart without starting a whole new thread

[twilit smelt]

displaying the name and boot badge of the booted OS on the sidebar

yea or nay

[mortal barn]

looks pretty cool actually

[cursive charm]

yea

[twilit smelt]

need to make little bitmaps to represent CPUs now

@icy bridge i made a tux boot badge bitmap for you

[icy bridge]

oo thanks

[twilit smelt]

This Could Be You! (jerry rigged to display tux bitmap and say linux)

[dense vigil]

Lol

[dense vigil]

Extra points if it says the kernel version below

[keen stone]

why does it look so sexy

[twilit smelt]

it gets the os name from a hard coded string field in the boot block

so thatd be hard

itd have to be in there

[dense vigil]

hm

i guess it can only be pre-baked then

[twilit smelt]

i have to make a 40x64 CPU icon now

[twilit smelt]

40x64 cpu icon got hands

[rare epoch]

out of curiosity is this project really NT inspired?

[twilit smelt]

no i hate nt

[rare epoch]

banned

hating NT should be a bannable offense

[cursive charm]

i love NT when i don't have to think about win32

[twilit smelt]

this is the little cpu icon i settled on

pointing out that its a risc cpu is part of the 1989 larp

they couldnt stop saying risc back then

they stay grayed out except processor 0, to indicate they r idle

until they are IPI'd for the first time and then they color themselves in

to indicate active

the code for this sidebar stuff is some of the gnarliest ive ever written

only two things left

1. new boot picker
2. XRSI logo needs to be designed and put somewhere

but rn i need to take a break to get a bit of homework done

well ill do the boot picker first so that it reaches functional parity with the old gui and i can merge it

[rare epoch]

firstly that UI is so nice

secondly this feels like NT

[dense vigil]

Thats pretty neat

[dense vigil]

Did they actually have numa nodes back then and were they really that tiny

[twilit smelt]

the low ram is a quirk of the default ram being 4mb and the emulator splitting it evenly between the nodes (of which there must be 4 because 16 procs were specified)

[dense vigil]

Also why risc cpu and not the actual xr cpu name

[twilit smelt]

this is not a sane config lol

[dense vigil]

Oh ok

[twilit smelt]

its just a result of me putting -cpus 16 at the command line without putting a proportional amount of ram

they did have numa nodes

one of the AViiON systems was a 16 processor m88k machine with two numa nodes each with 8 processors i believe

[dense vigil]

And how much ram

[twilit smelt]

that was 1992

[twilit smelt]

a lot

probably like up to 1gb

(a lot for the time)

[dense vigil]

Yeah

[dense vigil]

What about this

[twilit smelt]

bc XR already appears too many places on that screen

[twilit smelt]

and also

[twilit smelt]

.

[dense vigil]

Oh ok

[shadow ridge]

risc of rain 2

[dense vigil]

Because u did have a prettier pic on the readme

[shadow ridge]

this firmware affects mintia 1 too right

[twilit smelt]

?

[twilit smelt]

wdym

[shadow ridge]

like you use the same firmware

to boot both

in the emulator

[twilit smelt]

this firmware has to chain load the old firmware to boot mintia1

[dense vigil]

U have an xr cpu pic on the readme somewhere

[shadow ridge]

oh

isnt that like

too hacky

[dense vigil]

Backport mintia1 to new firmware

[shadow ridge]

^

[twilit smelt]

no its way simpler lol

[shadow ridge]

yeah but they wouldnt do shit like this in the 80s

or would they

[twilit smelt]

yes they would

[rare epoch]

I think that’s nicer and not hacky

[twilit smelt]

that literally is how NT would boot on some systems

[rare epoch]

it reminds me of third party CSM modules

[shadow ridge]

i saw that

[rare epoch]

yes autocorrect

[twilit smelt]

also mintia1 is not the only OS that relies on the old firmware

theres also aisix

it chain loads it to boot both of those

[shadow ridge]

why tf did it auto correct to Jacky 😭 😭 😭

[rare epoch]

idk
maybe it thought it was a name

[warm mural]

aisix revival when

[rare epoch]

AIsix

AI OS

[twilit smelt]

never it was awful

[shadow ridge]

so theres the XRfirmware

that chainloads LIMNfirmware

that boots aisix

[warm mural]

No it was great

Way better than mintia2 even

[shadow ridge]

and all of this runs in computercraft

is my understanding correct

[twilit smelt]

new firmware gui is up on the web demo https://xrarch.github.io/

[twilit smelt]

yeah it can

i ahvent tested that in forever though

every time i test it i have to update the extremely ancient lua emulator with whatever platform changes occurred

[rare epoch]

Wtf?????

[shadow ridge]

ok what the hell

[rare epoch]

Isn’t computercraft like a Minecraft mod that lets you run Lua

[shadow ridge]

yes it is

[twilit smelt]

yeah

[rare epoch]

how can it run an OS

[shadow ridge]

you make it run the emulator

thats how hyena got into this shit iirc

[rare epoch]

😭

[twilit smelt]

this project started when i was a preteen on the computercraft forums ppl kept saying "its impossible to write a real OS for computercraft. you can only write a replacement shell." and i decided to prove them wrong by writing an emulator for a custom ISA, in lua ofc, and then writing a "real OS" for that

and it just kept growing until it outgrew computercraft

so i ported the lua emulator to use love2d

and then that got too slow eventually

so i rewrote the lua emulator as the C emulator

and thats where we are now

11 odd years later

[shadow ridge]

thats like more than a decade

[twilit smelt]

if you boot aisix or mintia1 u can see the old firmware come up lol

the one-to-one rewrite of mintia1 in jackal that i was working on was going to boot under a4x directly

but it started to dawn on me how vast mintia1 was and that that was going to take me months so i gave up

directly rewriting 120k lines isnt a cake walk

[shadow ridge]

The MINTIA operating system provides an environment suitable for use in a range of systems, from single-user desktop workstations to minicomputers supporting dozens of concurrent users. It does this while retaining performance and reliability.

shouldnt it be the other way around

[twilit smelt]

?

[warm mural]

minicomputers are big despite the name

the pdp11 was a minicomputer

[twilit smelt]

yeah minicomputers are the size of fridges

[shadow ridge]

oh

[twilit smelt]

hence the term microcomputer

smaller than a minicomputer

your pc is a microcomputer

[shadow ridge]

is my phone a femtocomputer then

or nanocomputer maybe

lol

[twilit smelt]

80s terminology:
computer = size of a room
minicomputer = size of a fridge
microcomputer = fits on a desk

[shadow ridge]

so what would a single user desktop workstation be

a minicomputer

or like a computer

[twilit smelt]

a microcomputer

[shadow ridge]

okay right

wait aisix is a unix

[twilit smelt]

only if you stretch the definition of unix

it was my first semi-viable preemptive multitasking os project ever

2020

[shadow ridge]

it takes forever to compute fibonacci sequences

[twilit smelt]

its recursive fib

non memoized

so yeah

[shadow ridge]

where is the page file on aisix

i cant find it

[twilit smelt]

it did not swap

includes the earliest comment documentation of my attitude toward swapping

17 yr old me wrote that

godspeed to him

[warm mural]

Did you take this comment syntax from OCaml

I've never seen it anywhere else

[twilit smelt]

pascal

[delicate dome]

the 1980 RISC LARP is so cool

[twilit smelt]

1990*

[delicate dome]

oh

i see

How far is MINTIA2 progressed, anyway?

I know old MINTIA had a whole userspace running right?

[twilit smelt]

this is what a crash look like now btw

[twilit smelt]

not as far as i would like

[delicate dome]

What was RAM limit again?

[twilit smelt]

@icy bridge this is all pushed so if you wanna try to get linux looking good its stable now

[delicate dome]

16 MB or smth?

[twilit smelt]

256MB

thats for a fully loaded "xr/frame"

[delicate dome]

That number's definitely higher than last time I asked

[twilit smelt]

xr/station is up to 32mb for 1 node

its not

its been 256mb since 2018

so thats unlikely

its 4mb by default

4mb 1 processor would be a typical config youd come across

[rare epoch]

are the CPUs 32-bit or 64-bit

[twilit smelt]

32 bit

[delicate dome]

What kind of virtual memory does this do? Radix tree paging?

[twilit smelt]

software tlb refill

[delicate dome]

Oh interesting

So like MIPS where software decides what page maps are

That makes the CPU a lot easier

[short owl]

I still don't get how these things work lol

[delicate dome]

Basically, instead of having hardware do virtual to physical translations, software does it. And then just stores said translations in the TLB (Translation Lookaside Buffer; cache of translations) explicitly.

[short owl]

Ah

So the software is the one that determines the page tables

[delicate dome]

yes

[short owl]

Well that definitely is easier than making a hardware page table walker

[delicate dome]

yes

but less efficient

[short owl]

I mean that's why we have the TLB

[delicate dome]

No I mean

Both have the TLB

But the mere fact software is bothered to fill it, makes it way slower than a hardware page table walker would be

[twilit smelt]

That wasn't always true

The software refill on many risc chips was famously faster than the hardware refill on x86 for some time

Partly because x86's walker couldn't take advantage of software optimizations

[delicate dome]

i.e. describing mappings in a more compact manner in the TLBs?

causing fewer faults?

[short owl]

Hmmm, and what happens when you try to access a page that isn't on the TLB at the time, does the cpu raise a fault so the software fills some entry?

[twilit smelt]

Yes

[twilit smelt]

Bottom up page table walking was a common one

[delicate dome]

Well what I can confidently say is that on modern CPUs hardware walkers are everywhere for a reason

[delicate dome]

google has become literally useless... what is that?

[twilit smelt]

They have enough silicon to do multi level TLBs and stuff which essentially does bottom up page table walking in hardware

[twilit smelt]

It's where when you take a tlb miss exception, you try to grab the PTE directly from a recursive mapping of the page tables. If the containing page table is in the TLB already, then you get the PTE right away and don't need to walk the rest of the levels. Otherwise you take another TLB miss on it and that time you end up looking at the PDE, and so forth until you get to the top level table which is anchored in place with a "wired" TLB entry that is never replaced

This works bc of how recursive mapping works which I won't explain

if u don't know alrdy

but basically this saves you from having to look up the upper levels of the page tables

most of the time

[delicate dome]

I know recursive mapping

[rare epoch]

Software TLB feels ugly

I think caches like this should be mostly transparent to user

[mortal thunder]

They are

User mode never sees such caches

[rare epoch]

user was bad phrasing, I meant osdever

[rare epoch]

out of curiosity does software TLB mean the OS can implement its page tables in any format it desires?

[dense vigil]

Yes

[rare epoch]

cool tbh

[icy bridge]

yeah but there's usually no reason to use anything other than the CPU's tlb entry format

[shadow ridge]

you dont even need page tables you can use any data structure and get away with it

(well by page table i mean radix tree)

[twilit smelt]

i think itll look better if it goes on the right

it feels imbalanced to have the console text and the sidebar stuff all putting so much visual weight on the left

[twilit smelt]

"user" in the context of like an architecture manual refers to the osdev

who is using the chip

[warm pine]

sidebars traditionally go on the right

i have to admit the idea is quite a nice one

and since all monitors are widescreens today i might or might not copy the sidebar idea for the keyronex console

but i'd add an analogue clock, that would add a classy touch, like a lexus dashboard

[shadow ridge]

what if the TLB refill routine used a random number generator to decide the entries' page frame addresses

[short owl]

Btw is it the firmware or the kernel itself the one that starts the other cores?

[shadow ridge]

why would the firmware start the other cores

[twilit smelt]

in this platform they all start simultaneously at reset, processor id 0 goes on to become the boot processor and the others enter a hlt loop waiting for an IPI

the kernel IPIs them (using a firmware interface) to get them to start executing kernel code

[shadow ridge]

how do you decide the start PC then for each core

[twilit smelt]

they all start at 0xFFFE1000

which is part of the boot rom

[shadow ridge]

so the code there does a check like if(WHAMI != 0) while(true) hlt();

[twilit smelt]

FwxReset:
.global FwxReset
    // Zero out RS.

    mtcr rs, zero

    // Invalidate the caches.

    li   t0, 3
    mtcr icachectrl, t0
    mtcr dcachectrl, t0

    // Set our exception block.

    lui  t0, zero, 0xFFFE0000
    mtcr eb, t0

    // If we aren't processor zero, go to the MP corrall and wait for an IPI.

    mfcr t0, whami
    bne  t0, .mp_corrall

    // Reset EBUS to quiescent state.

    la   t0, 0xAABBCCDD
    mov  long [0xF8800000], t0, tmp=t1

    // Set the initial stack pointer to 1024 bytes.

    li   sp, 0x400

    j    FwReset

.mp_corrall:
    // Set the stack pointer to 0x400 + (320 * id).

    li   sp, 0x400
    add  sp, sp, t0 LSH 8 //  + 256 * id
    add  sp, sp, t0 LSH 6 //  + 64 * id

    mov  a0, t0
    j    FwProcessorStartup

[short owl]

Because it shows you in the sidebar which cores are started, so i suppose it does have an involvement on that

[twilit smelt]

thats the reset code

[twilit smelt]

yeah it updates the icon when the OS calls the firmware interface to IPI them

[shadow ridge]

ok that kinda makes sense

[twilit smelt]

actually it has the processors update the icon before they jump to the OS-specified entrypoint

so it only gets updated if the processor itself woke up correctly

[shadow ridge]

what is EBUS btw

[twilit smelt]

the system bus

[shadow ridge]

and what does it do

[short owl]

Busing around

[twilit smelt]

from the system software perspective it provides a general layout for the physical address space

from the fantasy hardware perspective i suppose it would be a complex protocol of electrical signals over copper traces and a slot form factor and so on

(bus)

[shadow ridge]

hmm

for a second i thought it was like the equivalent of PCI

for your thing

[twilit smelt]

i mean it is

its less fancy but thats basically what it is

[shadow ridge]

oh

if they make a C compiler for limnstation ill port my kernel to it

[twilit smelt]

there is one

monkuous ported linux to it using that

[shadow ridge]

holy shit

it boots?

[twilit smelt]

did you not know that

[shadow ridge]

no

or i forgot maybe

[twilit smelt]

boots and runs bash and glibc

[shadow ridge]

shouldve ported neofetch

thats incredibly cool ngl

why does it say This architecture does not have kernel memory protection

[twilit smelt]

idk

i think it means monkuous compiled it with kpti disabled

or something

the coolest thing about it for me is that it is the ultimate vindication that my fake computer is a real computer

real enough to boot linux

and also its gratifying to see linux running atop my tlb refill scheme

[shadow ridge]

this also means your emulator doesnt have like

any major bugs that linux could trigger

[twilit smelt]

i mean last time monkuous worked on it linux did crash randomly but he thought that was more likely his fault than mine

[shadow ridge]

oh

[twilit smelt]

its probaly not my fault because mintia1 should have exercised the probably related emulator components like cache and mmu enough to weed out any bugs linux could be triggering

[warm mural]

btw I think it looks way better this way

and the borders are much nicer than before

[mortal thunder]

i cant wait until you start working on a window system for mintia

i know its like

a couple years away probably

(pessimistically speaking)

but i can't imagine it looking anything less than great

[twilit smelt]

Someone enlighten me on the state of the art in slabs and magazines

[rare epoch]

the two words I don’t want to hear again

[short owl]

Bonwick's paper?

[twilit smelt]

I've read it

That's not the state of the art

That's the like origin of the art

[short owl]

Which one is it then

[twilit smelt]

I've been checked out somewhat while @warm pine and @warm mural have been studying it and have missed out on most of their cool findings

Ik abbix has a cool bitmap thing

[dense vigil]

#1485107693458030706 message

[warm pine]

This is trivia but Linux has implemented them for a second time but renamed them

They implemented them ages ago in what they called their "SLAB" allocator, then someone said "SLAB has too many queues and too much per CPU, let's replace it with SLUB which won't have either", then SLAB disappeared, and now magazines have been reimplemented exactly per Bonwick and Adams

[dense vigil]

there's also SLOB for embedded systems

[warm pine]

In a cryptic move they've called their "magazines" and "depots" as "sheaves" and "barns" respectively

[warm mural]

Like freebsd

It has kegs and buckets

[warm pine]

Kegs being a reference to alcohol and buckets to marijuana

The analogies are always so interesting

[raven drift]

i don't get this desire to rename existing concepts

In my experience, this is usually a sign that the author of the code does not know the state of the art very well

[warm mural]

I renamed caches to zones (as per freebsd naming) because it makes more sense imo

[coarse current]

There's also SLOP now
(openslopware)

[rare epoch]

you mean like Limine?

[coarse current]

yeah

[warm pine]

The author did not even refer to magazines in his patch submission which was bizarre

[rare epoch]

maybe they rediscovered it somehow?

[warm pine]

He only later when quizzed about the name revealed he had implemented them as per bonwick and Adams' 2001 paper

[rare epoch]

oh

tbh I don’t like the names in the original paper

too fancy

[warm pine]

I forgot why or if he said why he changed the name. But I think he said somewhere that magazine mighy be evocative of rifles, which is culturally insensitive to the people of the USA, where they have numerous school shootings as well as many other kinds of shootings

In USA they have strong feelings about whether these shootings should contikue and - you guessed it - if you have a difference of opinion with your interlocutor, it could end in tears and a bullet (just ask Charlie Kirk) But it still feels extreme to purge any word that can be associated with firearms from the kernel

[twilit smelt]

Crazy Charlie Kirk came within 10 miles of me and that happened to him right after

[dense vigil]

wdym by came

[shadow ridge]

he was in utah

[dense vigil]

oh

[shadow ridge]

they got that shooter for attempted murder right

okay nvm this joke is getting old

[twilit smelt]

Got an evacuation order bc they found a gigantic cache of explosives

[shadow ridge]

what the fuck

only in utah 😭

[dense vigil]

are u evacuating now

[twilit smelt]

No that's miles from me

[rare epoch]

what country?

oh AT&T

so USA

where?

[dense vigil]

where are you even supposed to go in this case

[twilit smelt]

Utah

[night needle]

he's with tung now

[raven drift]

well, outside of the affected region

[short owl]

https://tenor.com/view/tung-tung-tung-sahur-tung-tung-god-tung-brainrot-gif-1470230787957426753

[mortal thunder]

i find it funny how i can tell this is an iphone se 3rd gen from the top bar

also what does 2000 feet of 1100 W 1400 N mean

are people expected to know their precise geolocation in degrees on the meridians/parallels?

[craggy spire]

I'd guess those are street names

[twilit smelt]

In SLC yes

[mortal thunder]

lol

all i know is that the 45th parallel cuts through my country

middle, slightly south

[warm pine]

during the scottish enlightenment our town planners decided that it would be more rational to lay out towns and cities on a perfect grid

[shadow ridge]

south lake city. more like south 12.32 east 12.75

or something idk

[warm pine]

glasgow is one of the poster boys of this movement

the americans were very influenced by our ideas so they built their cities on grids

[mortal thunder]

ive seen that before

i thought the americans were the ones who started that trend

lmfao

[warm pine]

so you can just locate yourself by a coordinate relative to some centre point

[shadow ridge]

i wonder what the most efficient road layout would be for a given amount of buildings and the average traffic for each one in a city

[warm pine]

the americans like to film in glasgow when they want to depict "new york city but 100 years ago"

aberdeen has a more ignoble claim to fame, they filmed some scenes here which were intended to be set in the late soviet union!

[mortal thunder]

i feel like if let loose our former communist regime would've done something similar

[twilit ingot]

i very much hate this

[warm pine]

there is an apple film about tetris and you'll find the headquarters of the firm is in fact the zoology building of the university of aberdeen

[mortal thunder]

but ceaushima wasn't so bad as to warrant redrawing streets

[shadow ridge]

like a grid doesnt really sound efficient for traffic

[twilit ingot]

it's not

[shadow ridge]

but it's just a feeling idk how to prove it

[twilit ingot]

some guy on yt did a vid about it

[shadow ridge]

ok maybe a start would be that it has like N^2 intersections

[warm pine]

it's said that cumbernauld in scotland or milton keynes in england may be some of the most efficient designs

[shadow ridge]

yeah but is it mathematically provable to be the most efficient

[warm pine]

milton keynes is grids within grids but it's really the separation of pedestrian from motor traffic that works for both of them

[warm pine]

maybe it's the subjective experience that matters more

[shadow ridge]

idk i dont trust human judgement on time perception

if you stopped whatever you're doing and just stared at a wall for 1 minute each day, that would feel worse than the same thing but done for a single second at a time, 60 times a day

[warm pine]

that's maybe the trick of it

[warm mural]

I don't think this is even a solvable problem

[warm pine]

in any case living in cumbernauld probably feels a bit like doing that 24 hours a day

[shadow ridge]

i know this place

it's from silent hill 2

you cant fool me

[warm mural]

Looks like Harry potter

His house

A bit

[shadow ridge]

the houses there actually look good tho

[warm mural]

Yeah true

[shadow ridge]

this looks like brutalist crap

[warm pine]

some other districts of the town

[twilit ingot]

i also very much hate this

[shadow ridge]

i get why they would want to film soviet movies in places like these

[twilit ingot]

all the houses are equal and grey

bruhh

it's hell

[shadow ridge]

these were probably built many decades ago

when brutalism was a thing

because people were like "wow what if we built like this, it would be so cool!" until they actually built it and realized it sucks ass

[warm pine]

it was the fashion until the 70s or so

milton keynes was built after that so it's much nicer

despite also being a town built practically out of nothing

[shadow ridge]

why not

[warm pine]

it certainly seems like it would be difficult to model

and you have so many things to take into consideration

for example do you care how congested it gets if there's a football match? are you prepared to figure out how you ought to account for traffic from fans both home and away? what about other events?

[shadow ridge]

isnt it like any other algorithm

like even if the worst case is O(N^3) or whatever

you can still have an average O(N) ig

idk what that would mean for a road though

maybe you could always design for the worst case or design the roads differently for both the average and the worst

like making longer routes around the city for big events that have more capacity

idk im not a civil engineer

[warm mural]

In most cases brutalism wasn't chosen for it's aesthetic

it's just cheaper to build with concrete

[shadow ridge]

it's still cheaper, no?

you're still better off using basic geometric shapes and not doing any sort of outside facade work

money wise

and it's not like they were in the great depression either

[warm mural]

that is literally what I said?

you were talking about brutalism as if people chose to do it because it looked good

[twilit smelt]

brutolesm

[warm mural]

lol remember brutal

[twilit smelt]

yeah they kept bragging about how they were 50k loc

remember when that was a lot

the osdev power scaling sure has gotten out of hand

[delicate dome]

50kloc isn't that huge, is it?

[warm mural]

idk if its even 50kloc

[delicate dome]

I know Managarm is way more

[warm mural]

having worked on it I highly doubt brutal is 50kloc

[delicate dome]

BadgerOS is about 50kloc ignoring headers

[warm mural]

nvm wtf

how

ah a lot of lines are from the math library

[delicate dome]

Oh that's funny

BadgerOS has gotton smaller in terms of loc

[warm mural]

and 50kloc is quite a lot

imo

at least for a solo project

[rare epoch]

it is yeah

my current kernel is still like 5k loc

[delicate dome]

damn

Guess I'm a certified masochist after all

[warm mural]

mine is ~11kloc right now and it doesnt do much

[night needle]

huh

[delicate dome]

I was curious and Boa32, my RISC-V CPU, is 20kloc

[shadow ridge]

afaik that was a big factor

[shadow ridge]

to me it seems we would all still be building cities in a brutalist style if we went by that logic

[shadow ridge]

my kernel is only 1.1 kloc as of now

which makes me realize i should work on it more

[twilit smelt]

how

[delicate dome]

Well that includes shitloads of peripherals and test code

But same project

I wonder where all that asm comes from because I did not write that much

CPU itself is 4kloc

peripherals another 4.4k

[delicate dome]

Found it

It was counting the submodule riscv-tests (official RISC-V unit tests) as part of my project even though it isn't

[mortal thunder]

if i count absolutely everything i can get pretty close ```/mnt/c/boron$ cloc --vcs=git .
552 text files.
529 unique files.
62 files ignored.

github.com/AlDanial/cloc v 1.90 T=13.76 s (36.7 files/s, 5232.0 lines/s)

Language files blank comment code

C 256 9366 6974 32605
C/C++ Header 176 3171 4143 10784
Assembly 16 321 500 1539
make 33 197 210 554
Markdown 4 169 0 483
C++ 6 70 45 317
Pascal 2 8 0 160
DOS Batch 3 26 5 116
Python 2 46 24 111
Bourne Shell 7 6 4 37

SUM: 505 13380 11905 46706
-------------------------------------------------------------------------------```

but theres a lot of external code and things

and the kernel is only about this many```
/mnt/c/boron/boron$ cloc --vcs=git .
279 text files.
278 unique files.
11 files ignored.

github.com/AlDanial/cloc v 1.90 T=1.83 s (146.1 files/s, 23908.4 lines/s)

Language files blank comment code

C 149 6130 4987 20402
C/C++ Header 101 2009 2707 4811
Assembly 12 286 463 1368
make 1 55 47 175
Pascal 2 8 0 160
Python 2 46 24 111
Markdown 1 12 0 44

SUM: 268 8546 8228 27071
-------------------------------------------------------------------------------```

also your directory structure is kinda fucked, why the hell are you using single character directory names

r/z/s/zinnia

just put them in like ~/src or ~/work or ~/projects

[night needle]

what

that's fish doing it

it's repos/zinnia/sources/zinnia

[mortal thunder]

what is fish and why does it do that

[night needle]

best shell

it does it so the path in your prompt isn't too long

[mortal thunder]

aha, so a path compression of sorts

imo if the path is so long that you need a shell to compress it you have bigger problems

[night needle]

not really

it always does it

it's just 4 components here

[acoustic sparrow]

i mean personally id just make the path shorter by removing useless crap but sure i guess thats one strategy

[night needle]

yeah but i think it's already as minimal as it can get

[acoustic sparrow]

idk man its not very hard to make to make it more minimal

[night needle]

it's essentially projects/distro/sources/kernel

[acoustic sparrow]

the whole "(wayland)" part is just noise tho

[night needle]

that's the branch

it's useful info

[acoustic sparrow]

is that really that important lol

[night needle]

it helped me more than it has caused issues

[acoustic sparrow]

i see

the only thing like that in my prompt i actually have is the active direnv tbh

[night needle]

direnv?

[acoustic sparrow]

its this thing: https://direnv.net/

[night needle]

ah

[acoustic sparrow]

its useful if you are doing nix because it means you dont have to use their shitty bash setup

you just echo use nix > .envrc, direnv allow, and its all in path now

[night needle]

i like fish because it has really good completions most of the time

[acoustic sparrow]

fair

zsh has some weird bugs/issues (probably caused by my bad prompt) with completions

[warm mural]

I used fish for a while and I think fish is pretty good but its not POSIX

[night needle]

yeah tbh that's its biggest downside

[warm mural]

I think zsh with oh my zsh gets pretty close to its functionality while still being POSIX

[acoustic sparrow]

i dont think zsh is posix?

[cursive charm]

it's POSIX-compatible, i'm pretty sure

[acoustic sparrow]

no its not

[warm mural]

it's closer at least

[shadow ridge]

i just use bash

[twilit smelt]

im like gon freeccs if he were into hobby osev

[twilit smelt]

His head is remarkably flat

Especially with the ears going to the side

Enhances the flat appearance

Possibly a 99th percentile flat cat or more

[shadow ridge]

does he have a flat brain as well

[blissful smelt]

have you tried out tokei

it's very fast and automatically recognizies .gitignore

it takes 4 seconds to count the linux kernel's loc compared to 4 minutes for cloc

[warm mural]

scc is pretty good too

i think tokei is pretty much just scc but rust

[mortal thunder]

No I haven't

[twilit smelt]

I came extremely close to failing my algorithms final just now

There were 4 questions and 2 hours

After an hour I'd answered two and was completely stumped on the others

I stared blankly at them with my panic increasing for like 45 minutes

Then for one of them I was suddenly like oh the solution is just a fucking knapsack problem with an extra loop at the end

Quickly scribbled it down

Looked at the other one

Stared at it and suddenly had a brain blast from a homework problem from two months ago and scribbled it down

Finished with seconds remaining

From what I overheard from other students they did NOT fare as well mostly

[shadow ridge]

shouldve just asked claude

[twilit smelt]

https://www.youtube.com/watch?v=oaI2Db3WBoE

[twilit smelt]

i'm going to design a kind of hierarchical bitmap where the lowest level is a bitmap and the higher levels are bloom filters

and put it in Rtl as a general utility

itll be designed for each level to be virtually sparse so that they can be extended

[twilit smelt]

also sometimes i wish i kept the mintia source code private

not for any good reason

just to like make people constantly question whether its real

[shadow ridge]

the first mintia was basically as good as private anyways

like if i encrypted my code and put it on github

would you say it's public

[twilit smelt]

monkuous could read it

w/o any help from me at all

[dense vigil]

Monkuous can read encrypted stuff as well

[shadow ridge]

he can probably break RSA in his head

pure neural power

[twilit smelt]

fadanoid could also read it

[mortal thunder]

you wouldn't be nearly as popular in this discord if the source code was kept private probably

people would doubt your thing is even real and exists

[twilit smelt]

also mintia2 is essentially encrypted too because its reached such a high level of esotericism

if i was a beginner and i was like "ooh this seems like an interesting hobby project to learn from" and it was mintia2 and i looked at that id probably just quit

[twilit smelt]

this is my current "play on loop infinitely while doing osdev until i hate it" song

found it in a mixtape made by boards of canada

called societas x tape

best song in a p good mixtape

[hybrid condor]

@twilit smelt I think you're the right person to ask

do you know of any reliable resources on the internal workings of the ExAllocatePool3 allocator in modern windows?

[twilit smelt]

you may be saddened to know that i dont even know what that API is

but i can direct you to @hollow musk or @craggy spire who might

also kelowerirql in my discord

[dense vigil]

probably ida

[hybrid condor]

unfortunately I have to rely on 3rd-party descriptions

[dense vigil]

wdym have to

[twilit smelt]

look at their bio

[dense vigil]

whats it supposed to tell me

[twilit smelt]

that im fat

[dense vigil]

is that like a slang for something

[twilit smelt]

no thats just my filler reply when i dont have anything else to say

[dense vigil]

nah i was replying to your bio message

[twilit smelt]

i said look at THEIR bio

not my bio

[dense vigil]

I DID 😭

[digital pivot]

will there be ExAllocatePool4 one day ?

[raven drift]

that they're not allowed to use RE, probably

[mortal thunder]

It just says "high paid plumber" so probably not

[acoustic sparrow]

U+F8FF is used by apple to represent the apple logo

[acoustic sparrow]

which scenario of it?

afaict exallocatepool3 can give you memory from a normal pool or from the secure pool

my guess is that its something like:

• on first use (or some initialization i cant find in the docs), the pool state associated with the tag is allocated and stored in a hashtable
• then allocations proceed as normal with the given {pool, size} pair
• secure allocations work by doing everything in the secure kernel through a proxy

but i havent read the code

[acoustic sparrow]

or the tag and cookie may be stored before the allocation

[acoustic sparrow]

fwiw if you can read first party microsoft blog posts: https://www.microsoft.com/en-us/security/blog/2020/07/08/introducing-kernel-data-protection-a-new-platform-security-technology-for-preventing-data-corruption/

[acoustic sparrow]

(3) is confirmed by the blog post so yeah

the internal RtlpHpAllocateHeap function is basically just void* RtlpHpAllocateHeap(heap: *SegmentHeap, size: usize, flags: usize, unknown: u16) according to this pdf from ibm security

[mortal thunder]

what good is that option if everybody who doesn't use apple software can't see it

[twilit smelt]

i like how nobody thought that the weird unrendered character in their bio, might have anything to do with the apparently confusing statement i wrote about their bio, and so they didnt google it

so they were really loudly confused and my attempt to be discreet backfired catastrophically

[acoustic sparrow]

sorry

but yeah lol

[craggy spire]

It uses the same allocator under the hood as ExAllocatePool2 or any other pool API

A key feature of ExAllocatePool3 is the ability to specify additional flags such as the aforementioned Secure Pool

However, one of the latest kernel updates this year removed Secure Pool, and its allocations now return a not supported status

26H2 and later switch to a new pool instead of Secure Pool, which isn't even a proper pool in the usual sense for NT

[balmy zealot]

What sort of algorithm is a pool allocator?

[digital pivot]

iirc it uses lock free linked lists

[twilit smelt]

for caches yeah

[shadow ridge]

utf standard says it's in a private use area

[dense vigil]

what does it mean

[warm mural]

just google it 😭

[dense vigil]

https://www.compart.com/en/unicode/U+FBFF

and

ah wait thats 8

https://www.compart.com/en/unicode/U+F8FF

so tldr they're an apple employee

[dense vigil]

how do u know all of this stuff

[balmy zealot]

That is to say, a pool allocator differs in that upon deallocation it caches the memory block in a free list?

[twilit smelt]

pool is just another word for heap

that is an optimization that the windows kernel's heap/pool allocator just happens to perform

theres no special thing called a pool

[balmy zealot]

For some reason, out of unfamiliarity, I assumed it was the name of some clever 'pool' algorithm.

What algorithm actually underpins a pool allocator?

[twilit smelt]

there are many

everyone does it differently

there are many ways to divvy up memory into blocks to hand out and then making it more scalable adds a whole other dimension of diversity on top of that

[craggy spire]

If you're specifically referring to Windows, in newer versions of Windows it's a pool (aka heap) wrapper over a segmented heap allocator and that allocator is itself a stack of suballocators chosen by size and pool type

The segmented heap is the primary general purpose allocator in both user mode and kernel mode with some modifications

So the allocator stack is:
ExAllocatePool
-> ...
-> pool manager, node and heap selection
-> dynamic lookaside, when eligible
-> LFH sub allocator
-> VS sub allocator
-> Segment sub allocator
-> Large block sub allocator

And each sub allocator has its own way of doing things

[craggy spire]

I'm updating to the latest versions and seeing what's changed

[dense vigil]

In ida?

[craggy spire]

WinDbg, IDA, a collection of various scripts

[dense vigil]

Do you have your own kernel or do you just re stuff

[craggy spire]

No, I don't

[dense vigil]

Sad

[digital pivot]

right lookaside is the word for what i was thinking of

now i remembered it

[cursive charm]

i saw the apple logo symbol at first and didn't think much of it until somebody pointed it out, lol

[craggy spire]

Yeah, not too long ago they switched to dynamic lookasides which have up to 64 possible size buckets with a bitmap indicating active buckets, and now depth is adjusted per bucket as part of global bucket selection

[cursive charm]

do you know the fault-tolerant heap in user-mode? supposedly, that's necessary for things like civ4 to run due to bugs in their CRT's alloc.

[craggy spire]

I know it exists, but I don't know any more than what's in the official docs

[hybrid condor]

thanks

[hybrid condor]

so no more Dynamic KDP?

[craggy spire]

It's gone with windows 11 25H2

[acoustic sparrow]

huh

i mean it doesnt feel like a terribly robust mitigation

[craggy spire]

They've changed their approach and the replacement will be more like Static KDP

[cursive charm]

first time i heard about it was apparently in a reply to a tweet from the guy behind hyprland (yes, really)

[twilit smelt]

right wing griftrland

[cursive charm]

here's the OG tweet in question if you're curious

the selected text was what had it catch my attention at first, lol

that is, apparently, "very lenient" is supposed to refer to the fault-tolerant heap

[twilit smelt]

is it

that just sounds like him saying shit

[cursive charm]

that's what i assumed since someone else would bring up fault-tolerant heaps in the replies

[acoustic sparrow]

lol

[cursive charm]

otherwise, idk wtf he was referring to

[twilit smelt]

we with our tiny audiences of experts reading our little messages are held ~100x more accountable for intellectual honesty than tech influencers are

and that gets to them a lot

[acoustic sparrow]

i thought vaxry isnt an influencer?

[cursive charm]

NT is not any "more lenient" with mm than linux, so either he means fault-tolerant heaps or he doesn't understand what he's talking about

[acoustic sparrow]

hes the developer of hyprland

[twilit smelt]

influencer and developer arent mutually exclusive

[acoustic sparrow]

true

[twilit smelt]

influencer is just someone with a big audience, and they can also happen to have real substantial projects, but inevitably with a big audience of less-than-experts youre going to start saying silly shit about other things, it seems like a rule of nature

[hybrid condor]

LFH sub allocator looks like the front line, what is it?

[acoustic sparrow]

for smaller allocation sizes i think, yes

[warm pine]

i for one still don't know what this is

some sort of library that people use to write wayland compositors or something like that

[acoustic sparrow]

its a wayland compositor

no its just a compositor

which is very nih because he got banned from xdg iirc

[craggy spire]

LFH is used for small allocations once a size bucket becomes active, when it lacks memory, it obtains subsegments from the segment sub allocator

LFH is opportunistic, so if the bucket isn't active yet or if it can't provide a block allocation falls back to VS sub allocator

[cursive charm]

hyprland is a wayland compositor

[hybrid condor]

what is meant by activation?

[twilit smelt]

i love this part of his blog post about it where he says "excess online hate" implying that there was a goldilocks zone of online hate he was aiming to direct at them and his audience just went too far

you guys were only supposed to do 97 online hate units not 120

not cool

[acoustic sparrow]

lol

[cursive charm]

LMAO

vaxry isn't very smart, i guess. that much i know.

[craggy spire]

LFH uses lazy promotion of a small size class, so when a bucket is not enabled its bucket array entry records usage. When the same size is allocated often enough, the bucket becomes active and allocations of that size can go through LFH.

If I'm not mistaken, often enough means 17-18 times

[cursive charm]

i still don't know how he managed to say "windows is very lenient with memory management" when:

1. official docs for fault tolerant heaps would imply civ4 was crashing on NT before:

The Fault Tolerant Heap (FTH) is a subsystem of Windows 7 responsible for monitoring application crashes and autonomously applying mitigations to prevent future crashes on a per application basis.

2. heap corruption would almost certainly lead to a crash in most situations anyway

[warm pine]

anyone who likes vax can't be so bad

[cursive charm]

people like vaxry?

i just assumed he was tolerated somehow

[cursive charm]

btw, if anyone here is in the hyprland server, let me know if he posts what i just said on his twitter, lol

apparently, he posts whatever people say about him online

[cursive charm]

i never had a VAX machine :(

[short owl]

He posts the fictional anime characters to which he goons to

(not a joke btw)

[twilit smelt]

thats normal zoomer behavior

[short owl]

Do you also do that then ig

[cursive charm]

i'd be sad if i did that on any of my social medias, lol

[short owl]

The Amazing Digital Footprint

At least use an AD account

Yeah sure i love seeing anime tits as well as updates on the software i use

[twilit smelt]

no.

[cursive charm]

doesn't he have a @hyprwm account for the latter?

[short owl]

Kinda but he also posts stuff on his main acc

[twilit smelt]

online zoomers are weird especially the right wing ones

[cursive charm]

vaxry is apparently a Self-Proclaimed Centrist™, so i'd say he's even more weird than most right-wing zoomers

[cursive charm]

i wish i still had a screenshot of one of vaxry's most infamous quotes, lol

it was one of the most baffling things i've ever read

[cursive charm]

do people generally think NTFS was derived from HPFS? a friend of mine just said somewhere that "NTFS was a natural development from OS/2's HPFS," but my understanding of NT's development was that there was apparently a lot of debate as to whether NTFS should be made.

[twilit smelt]

NTFS was a derivative of VMS's ODS

the ppl who designed it didnt care about or like HPFS

[cursive charm]

those would be gary kimura and tom miller, correct?

[twilit smelt]

yeah

i think ODS was a little more primitive than NTFS like i dont know that it was using btrees

but NTFS didnt get that from HPFS it just got it from the filesystem research that was current

same common source that HPFS and HFS got it from

[cursive charm]

yeah, i assumed the presumed relation to HPFS had to do with them using B-trees and that, according to wikipedia anyway, they share the same MBR partition ID (0x07)

i guess it makes sense if you forget most of NT was designed by people from DEC, lol

also, i just learned this exists: https://betawiki.net/wiki/Windows_NT_3.51_build_1057.1_(IBM_PowerPC_Edition)