MINTIA (not vibecoded)

[random heath]

correct

[twilit smelt]

and the kernel binary shrunk by like 200 bytes lol

while going to see how the asserts looked on fox32 i got sidetracked by a random intermittent crash and it turned out to be bad initialization of the virtual tlb in the fox32 emulator https://github.com/fox32-arch/fox32/pull/30

i debugged this and sent a PR in literally maybe 15 minutes

lmfao

[random heath]

yeah, something like that

although if they do that and believe they're good it's unlikely they'll get somewhere...

[dense vigil]

Why is this a problem if present is false anyway? You just invalidate this page when u get a pf and youre good no?

[heady bobcat]

why would the binary size shrink for !BLD_CHK builds?

[twilit smelt]

it shrunk for checked builds i meant

the assert messages were more compact than the ad hoc messages i had before

[heady bobcat]

oh lol

[twilit smelt]

basically virtual page 0 would match with an empty tlb entry where present=false

the emulator overloads this with the present bit having been zero in the pte

so itd not look up the page tables and would just unconditionally page fault at that point

[dense vigil]

Yeah I get that

Just invalidate the page and retry?

[twilit smelt]

this is a bug in the emulator because the fox32 tlb is not supposed to have negative entries

its supposed to work like the x86 one

this is a negative entry in the fox32 tlb ie a bug

[dense vigil]

well ok if the arch explicitly doesnt work like that then sure

[twilit smelt]

this actually wouldnt fix it

"invalidating" would just create a new entry that matches on page 0 and present=false

[dense vigil]

Hm?

[twilit smelt]

youd have to invalidate by cycling the page out of the tlb completely

by accessing like 32 pages or however many entries there are in the tlb

void flush_single_page(uint32_t virtual_address) {
    uint32_t virtual_page = virtual_address & 0xFFFFF000;
    //printf("flushing single page %X\n", virtual_page);
    for (size_t i = 0; i < 64; i++) {
        if (mmu_tlb[i].virtual_page == virtual_page) {
            mmu_tlb[i].physical_address = 0;
            mmu_tlb[i].virtual_page = 0;
            mmu_tlb[i].present = false;
            mmu_tlb[i].rw = false;
            //printf("flushed\n");
            break;
        }
    }
}

[dense vigil]

Hmm, but your pr doesnt fix this bug entirely then?

[twilit smelt]

it does

because i also fixed this spot

[dense vigil]

Oh ok

[twilit smelt]

this would be a fun chip bug to work around if it were real

and then id have to do something like what you said

[dense vigil]

Why do you need the zeroth page anyway?

[twilit smelt]

and this

[twilit smelt]

bc i identity mapped the low 4mb of ram before jumping into the kernel and it accesses the IVT through the identity mapped zeroth page

during initialization

[dense vigil]

Damn

[twilit smelt]

this worked intermittently bc sometimes id have touched enough pages to have cycled out the empty entries and other times i didnt

i was lowkey just ignoring this crash before because itd magically go away if i rebuilt the whole kernel (i guess somehow affecting the number of pages touched between the loader enabling paging and the kernel initializing the IVT)

so i figured it was just some inscrutable bug in my build system i didnt want to figure out

[dense vigil]

Cant you just setup the ivt and stuff before you enable paging? It seems kinda unsafe to identity map zero anyway as you won't notice null derefs

[twilit smelt]

i destroy the identity mapping a bit later in kernel init

thats another possible workaround yeah, or just mapping it in some other virtual page

but at the end of the day it was an emulator bug so i fixed it and sent a PR rather than doing crazy workarounds in my kernel

[dense vigil]

But yeah this could very well be a chip errata

Not that big of a deal anyway

I generally try to avoid cases where zeroing a struct somehow implies non default state

Ykwim

[twilit smelt]

this bug has been in the fox32 emulator for probably 2 years

maybe 3

it wasnt noticed in old mintia bc the kernel was big enough by the time it got a fox32 port that the loader would 100% of the time touch enough pages (while copying the kernel into the higher half) to remove all the empty entries by the time old mintia kernel would do this same thing to initialize the IVT

[dense vigil]

Whats the point of the present bit in that code if as you say it doesnt cache non present ptes anyway?

[twilit smelt]

youd have to ask ryfox

to be fair to her its the first time she ever implemented a paging mmu in an emulator and it held up well for a long time other than this and a few other oddities over the yrs

[dense vigil]

Maybe that logic should be removed altogether

[twilit smelt]

potentially

i tried to make my pr non-invasive tho

[dense vigil]

Makes sense

[twilit smelt]

mintia2 on fox32 is now happily fireworking again

[coarse current]

just change tlb cache non-present entries

[twilit smelt]

xr17032 already does that

[coarse current]

it's fun

[twilit smelt]

the reason to do that is to eliminate a branch in a software tlb miss handler

basically you dont have to check the present bit on the pte before loading it into the tlb

you can just blindly load it and return

and let the hardware notice if it was invalid

[coarse current]

loongarch does that

[twilit smelt]

when the instruction is re-executed and it matches with a tlb entry with present=0 and then it does a page fault exception that time

[coarse current]

exactly what you've described

I had to adjust my kernel to not shit itself on stale invalid tlb entries

[twilit smelt]

rather than going like (pseudo-assembly)

andi tmp, pte_value, 1 // isolate valid bit
beq  tmp, zero, PageFaultHandler
load pte into tlb
rfe

[dense vigil]

What's the point of that? Either you fill it right there, or proceed to segfault, either way the entry is never used again?

[twilit smelt]

basically you can eliminate 2 instructions from the end of the psuedo tlb miss handler here

[coarse current]

I know, I've written a software tlb refiller

[coarse current]

With software refilled TLB, you get separate exceptions for refill and for invalid entry

[twilit smelt]

you remove checking for validity of the pte from the hot path of tlb miss handling

[coarse current]

(At least I think that's their rationale)

[twilit smelt]

according to john mashey (old unix guy and MIPS architect) they did it like that specifically to remove the branch instruction from the tlb miss handler

while designing MIPS in the early 80s they had a prospective tlb miss handler on a whiteboard and every time someone came up with an idea to make it shorter theyd implement it

and that was one idea

[dense vigil]

So like it's guaranteed to be present?

[twilit smelt]

your tlb miss handler loads the pte for the virtual page you missed on

and then it just chucks it into the tlb

without checking anything about it

[dense vigil]

Ohh

[twilit smelt]

the second time the faulting instruction is executed, it'll match on the new tlb entry and the hardware notices the valid bit is 0 and then it does a page fault exception that time

[dense vigil]

To be fair it could just auto page fault if you load some placeholder address?

[twilit smelt]

rather than a tlb miss exception

[twilit smelt]

wdym

[dense vigil]

Like for example consider 0 a non present page

[twilit smelt]

isnt that this exact thing except more work

[coarse current]

you could maybe pagefault if you loaded invalid entry into tlb

[dense vigil]

Well no tlb entry used in this case

[twilit smelt]

like immediately? thatd trample on exception state in an evil way

[dense vigil]

Well in a safe way

[coarse current]

it's a possibility

[twilit smelt]

if you dont use a tlb entry then the hardware cant check it and then you just added a branch back into the software handler

[coarse current]

you could probably work around that

[twilit smelt]

and you may as well have branched on the valid bit

[twilit smelt]

read my architecture manual to see the evil state machine necessary to page fault safely in a tlb miss handler

[coarse current]

but like you rarely hit stale entries anyway so it's not a big deal

[twilit smelt]

the tlb miss handling scheme in my toy arch requires being able to page fault on stuff

in the tlb miss handler

specifically page tables

[dense vigil]

Your tlb miss handler runs with paging on?

[twilit smelt]

yes

[dense vigil]

Thats truly evil

[twilit smelt]

not only that, it loads the pte out of a recursive page table mapping

and can take nested tlb misses

that essentially perform an implicit bottom-up page table walk

[dense vigil]

Did you do it like that just for funsies?

[coarse current]

that's cursed

[twilit smelt]

its because i think its extremely efficient

[coarse current]

i see where the obsession for recursive page tables comes from

[dense vigil]

How is it efficient to recursively page fault

[twilit smelt]

after you fill in tlb entries for each level of the page tables, a future bottom-up walk can shortcut most of the paging hierarchy

[coarse current]

you can also just cache lower page table levels

[twilit smelt]

thats what this does

[coarse current]

without the recursive page table ... things

like all decent modern CPUs do afaik

[dense vigil]

Did any real arches do it like that btw

[twilit smelt]

this uses the same tlb for caching the lookup for all levels of the page tables

this is my entire tlb miss handler:

BxDtbMissHandler:
    mfcr zero, dtbaddr
    mov  zero, long [zero]
    mtcr dtbpte, zero
    rfe

its probably the most code golfed software tlb miss scheme ever

[dense vigil]

This works because recursive paging?

[twilit smelt]

yeah this looks like it couldnt possibly do anything useful but it does

basically dtbaddr is a control reg you initialize the upper 10 bits of with the 4mb-aligned base virtual address of your recursive page table

[warm mural]

no

here's mine on x86:

TlbMissHandler:

[heady bobcat]

and the zero register is banked 🤯

[twilit smelt]

when a tlb miss occurs, the cpu automatically fills the lower 22 bits with the offset the pte is at inside the page table (which is calculated basically as just vpn << 2)

[dense vigil]

Here's mine:

[twilit smelt]

yeah the zero register becomes a usable GPR during tlb miss handlers

so you can use it as scratch

it always reads zero otherwise

ignores writes

[coarse current]

Loongarch is ```mipsasm
tlb_refill:
csrwr $t0, 0x8B
csrrd $t0, 0x1B
lddir $t0, $t0, 3
lddir $t0, $t0, 2
lddir $t0, $t0, 1
ldpte $t0, 0
ldpte $t0, 1
tlbfill
csrrd $t0, 0x8B
ertn

[twilit smelt]

actually writes succeed but asymptomatically since you cant read it back

(unless youre in a tlb miss handler)

[twilit smelt]

mov zero, long [zero] loads the PTE from the address it got from this process

if there is no tlb entry for the page table its within, a nested tlb miss exception occurs - this is just setting dtbaddr to the new PTE (which is really now a PDE but this is calculated identically) and branching back to the start

repeat until you hit the top level page table which is anchored with a permanent tlb entry you set up earlier

then you fill out the tlb from the top down and later misses that hit in one of those page tables short cut the upper levels

[dense vigil]

Btw isn't there an issue there where to fetch an address of a pte you get into a tlb miss handler which fetches from the same page directory which again recurses into the same tlb miss handler?

[twilit smelt]

read what i just said

[dense vigil]

Since you have paging on

[twilit smelt]

you use a wired page of the ITB (instruction translation buffer) to map the tlb miss handlers

(there are two, an ITB and DTB miss handler)

(which are identical other than using the CRs for their respective TLB)

[dense vigil]

Oh you have a permanent tlb entry

[twilit smelt]

yeah the low 4 entries are permanent

in both tlbs

[dense vigil]

How do you set them?

[heady bobcat]

why?

[dense vigil]

Page table and handler itself

[heady bobcat]

yeah, but cannot there be a unified TLB?

[twilit smelt]

theres a dtbindex control reg that increments automatically every time you write to dtbpte, which inserts an entry in the DTB at the current index and then increments it

so you get rough fifo

the index wraps back around to 4 once you get to the top which is how those entries are made permanent

but you can explicitly set dtbindex to one of the first 4 entries and then write into dtbpte to set it

[dense vigil]

Thats pretty smart

[heady bobcat]

is there an equivalent of ASID/PCID for this scheme? or is everything flushed on context switches?

[dense vigil]

Like this sounds like it would work, but isnt it super slow

[twilit smelt]

yeah i have ASIDs

[twilit smelt]

wdym

[dense vigil]

It creates like a ton of tlb traffic to handle a page fault

[twilit smelt]

"it sounds like it would work" it does work mf i have a virtual memory manager successfully running stress tests on this thing that most people in here couldnt dream of on their boring x86 kernel

[coarse current]

it sounds like a normal amount of traffic

[twilit smelt]

no

[dense vigil]

How many levels are page tables on your arch

[twilit smelt]

again each level of the page table is cached in the tlb

2 bc its 32 bit but this extends naturally to any number of levels

[coarse current]

wdym by boring kernels

[dense vigil]

Oh ok

And how many entries in the tlb

[coarse current]

are they public?

[twilit smelt]

one is a swap test that intentionally causes swap thrashing and another is a forkbomb that just forkbombs

it can forkbomb indefinitely w/o crashing the OS and still recover when u ^C

ive run it for days without crashing

thrashing swap and doing all kinds of vmm things that idk of any other hobby vmm that can do em

https://github.com/xrarch/mintia/blob/main/OS/Commands/stress/Tests.df

they r written in cthuluspeak though

[dense vigil]

Yeah probably very very few hobby osses here can survive a fork bomb, let alone for a few days

[twilit smelt]

theres like 32 in the DTB and 16 in the ITB i think

but its not specified architecturally

unlike mips

[coarse current]

how do you survive it?

like killing the processes?

[dense vigil]

By not having bugs

Lol

[twilit smelt]

u can see the first four (wired) entries in the ITB, only one of them is used which is mapping the exception block which contains the TB miss handlers (among other exception entrypoints)

[coarse current]

like Linux dies

from a fork bomb

[dense vigil]

Hm?

[coarse current]

(its userspace at least)

maybe not the kernel

[twilit smelt]

the DTB maps two wired pages, one is the page directory at its proper position in the recursive mapping, and the other is the per-cpu block which is mapped at 0xFFFFF000 on all cpus

[dense vigil]

It will just oom at worst

[coarse current]

yeah

[coarse current]

but still

[dense vigil]

But u can set the oom adj score to -1000, so it will never be killed

(The original forker)

[coarse current]

I don't get it

[dense vigil]

So it will keep on forking

You will be getting ooms yes

[coarse current]

What do you mean by surviving a fork bomb

[twilit smelt]

no i just start returning error statuses from process creation once they hit their resource limits or exhaust system resources

duh

[coarse current]

Ok

[dense vigil]

The kernel doesnt leak memory, nothing blows up because of an oom

Successfully recovers once stopped

[coarse current]

but then the fork bomb stops once you run out of memory

[twilit smelt]

yeah you can ^C and wait like 30 seconds for everything to kill itself (bc most of the processes are entirely swapped out so this causes lots of IO as they leave existence)

and then get ur shell back

[dense vigil]

You can restart it in a loop

[twilit smelt]

like nothing happened

[twilit smelt]

no it doesnt

basically for maximum thrashing i just have a process terminate when it receives an "out of resources" status

which creates like a broiling sea of processes constantly popping in and out of existence forever

[coarse current]

so SIGBUS

[twilit smelt]

and beating the shit out of like every codepath

no?

[dense vigil]

Also mintia doesnt even overcommit

[twilit smelt]

fn (StressTestF) ForkbombTest { -- ok }
    while (1)
        auto phandle
        auto thandle

        "forkbomb"
        ArgTable@@ // path
        0 // creationflags
        0 // creationparams
        ACCESS_OWNER_ALL // permissions
        "forkbomb" // name
        OSSpawn ok! phandle! thandle!

        if (ok@)
            ok@ OSStatusGetName "%s\n" Printf
        end else
            phandle@ OSClose drop
            thandle@ OSClose drop
        end
    end
end

[dense vigil]

Uhh no, just an errno lol

[twilit smelt]

actually i guess it doesnt terminate

it just prints the error message

it probably used to terminate

[dense vigil]

Thats probably bad because once you use all available memory they just spin

[twilit smelt]

it overcommits phys mem but not virtual mem

[dense vigil]

By overcommits you mean ram + page file?

[twilit smelt]

yeah im pretty sure i changed it at some point and forgot to change it back

[coarse current]

how do you even overcommit virtual mem?

[dense vigil]

U cant lol

[twilit smelt]

virtual memory here means the capacity available in ram+pagefile

and not virtual address space

[coarse current]

Discord wouldn't work on that

[twilit smelt]

it would

because it only reserves 1TB

[dense vigil]

Wait so what exactly do you overcommit

[twilit smelt]

and doesnt commit it

[coarse current]

It requeses 1TB of ram

[twilit smelt]

it wouldnt run on windows if it does that

bc windows also forbids virtual memory overcommit

it will not promise you memory it cant back with whats available in ram+swap space

it tries to extend the pagefile before failing a request but it wont even try that if its asked for 1TB at once

that exceeds some heuristic auto-failure limit by quite a lot probably lol

[dense vigil]

What does that even mean? Overcommiting virtual is a subset of overcommiting physical

[twilit smelt]

no it isnt

[twilit smelt]

.

[twilit smelt]

.

[dense vigil]

Yes

Still a subset

[twilit smelt]

so physical memory is a subset of virtual memory

its the virtual memory that is currently cached in ram

[dense vigil]

Yes

[twilit smelt]

you can overcommit physical memory by asking for more than exists in ram

[dense vigil]

Im confused

[twilit smelt]

you cannot overcommit virtual memory by asking for more than exists in ram+pagefile

in mintia

[dense vigil]

Whats the difference vs nt here

[twilit smelt]

no difference

i also extend the pagefile if someone asks for more than is currently available and only fail it if i cant

and stuff

[heady bobcat]

what is a common policy for shared memory accounting/overcommit?

e.g. assume that there is a lot of memory pressure, no free physical pages, no swappable pages, and someone tries to mmap a file. what to do?

[dense vigil]

So mintia overcommit behavior is same as nt?

[twilit smelt]

yes

[dense vigil]

Nvm then

[twilit smelt]

file cache pages have backing on disk so they dont factor into this

you can always write them back to their file and discard them at any time

so theres no need to account them at all

by virtue of existing on disk theres room for them

[dense vigil]

What about shared anon

[heady bobcat]

yup. but you can't fault them in if there are no free/freeable pages

[twilit smelt]

im unsure why thats supposed to be a special case

[twilit smelt]

so you swap out pages

this might evict other file cache pages or it might evict anonymous pages

[dense vigil]

What if there's nothing to evict

[twilit smelt]

uhh

then you had a wired memory leak

whcih is a bug

and the system hangs

this would break any system

[heady bobcat]

then you fail the mmap

probably

[twilit smelt]

no

[dense vigil]

Well you've already done the mmap at that point lol

[twilit smelt]

yeah

[dense vigil]

You just start evicting pages that have just been read from disk

Basically like running on a platform that's collapsing behind you and appearing in front

[twilit smelt]

each shared page just counts as 1 page

you only need 4kb (or insert pagesize here) for it in ram or pagefile so theres nothing special about it, you just account it as 1 page

[heady bobcat]

if two processes mmap the same page of the same file without faulting it in, is accounting done by keeping track of a 'virtual PTE'?

[twilit smelt]

what on earth do PTEs have to do with this

[sterile frost]

how do you swap out to disk if you have literally 0 available physical frames

you need zero copy disk io

right?

or do you just keep a reserve of pages for critical things like that

[twilit smelt]

the answer is obviously no, its literally just a counter of "committed pages" you increment when you create anonymous pages. so if you create a shared memory region of 2mb and dont actually fault on it then youve incremented that counter by 2mb/4kb pages

faulting on it and PTEs are irrelevant

[heady bobcat]

there needs to be some datastructure for process B to be able to see that process A has already mmaped that page, and so it shouldn't be accounted

[dense vigil]

Why do you need a copy if it’s like a page of data anyway?

[twilit smelt]

it doesnt need any actual existence whatsoever other than the logical existence of "it exists"

[twilit smelt]

the data structure is the common shared memory object you already have

if youve done a competent shared memory implementation

[dense vigil]

Linux has struct address_space for this

It has a list of vm areas that map that file

[sterile frost]

i mean sure, if you can just

ahci_write_sector(...);```

then who cares about allocating anything, but if you have a big ass i/o system above all of that it becomes a bit more complicated than that i think

[twilit smelt]

and you account for the pages when you create the shared memory region not when you mmap it

[dense vigil]

Why does it become more complicated? Like if a device needs bounce buffers?

[twilit smelt]

why would that mean that

[sterile frost]

idk m,y mental model is missing like everything

[twilit smelt]

even if youre using programmed IO you can still just read the bytes from the page and shove them out the port

[sterile frost]

idk how that shit works

i am trying to figure that out

well assuming you use port io, yeah

[dense vigil]

You tell the disk to read from phys X and write to block Y

[twilit smelt]

but yes you usually do say like "i have 4 reserved pages" and you block threads when there are only 4 page frames left rather than when there are 0

and you only allow the swapper related threads to allocate from the last 4

[sterile frost]

yeah that's what i was trying to figure out

i was wondering if you just keep a reserve for critical things like that

[twilit smelt]

not like "that"

because that makes no sense

but for other things yes

[warm pine]

for disks this is basically intrinsic to all HBAs since a long time

[heady bobcat]

How is accounting usually done for pages that are shared across resource limit context boundaries (e.g. rlimit, cgroups, etc.)?

[sterile frost]

yeah but you need to somehow pass that information through the IO system, assuming you can use it in a page swap out scenario, but i dont see why you couldn't

[twilit smelt]

the only thing ive identified that ill need it for in mintia2 is IO packet locations which im going to dynamically allocate during IO enqueue

[warm pine]

if there is no scatter-gather then you can still write on a block-by-block basis

[twilit smelt]

for me, the creator is charged with it

[sterile frost]

but making sure that there are always a few pages available makes sense

[twilit smelt]

the creator can disappear and thats fine because the resource limit block has its own lifetime with its own refcount thatll be upped by the existence of these objects

[heady bobcat]

ok, makes a lot of sense actually.

[sterile frost]

SG was my concern (in the 0 physical pages left case) since you need to allocate some structures to perform the IO

i guess the SG list can go on the stack but the IO system might need to allocate some intermediate stuff like IOPs or whatever

[dense vigil]

I mean it makes sense that the kernel would have a pool of pages only for itself

[twilit smelt]

you can just set a field in the thread struct for the swapper that says "i am memory privileged"

[dense vigil]

To use for stuff like that

[heady bobcat]

in this case, does the "page struct" (or page struct equivalent) keep a reference to the "creator" for decommit later on?

[twilit smelt]

no, the shared memory object does

[warm pine]

i generally preallocate the necessary resources

however in all but the worst cases you should be able to do block-by-block writeback without further allocation

[twilit smelt]

i do this as much as i can but one instance i found where i cant is if the pagefile IO fragments and i need to allocate more associated IO packets

i mean i could preallocate enough for the worst case where it fragments on every single sector of the IO, and the max pagefile IO size is 64kb, and the max number inflight is 4, so then thats 4*64kb = 256kb, / 512 = 512 preallocated packets

at that point youre reserving enough memory you may as well just reserve some pages and make it a general mechanism for other things that might need it

[sterile frost]

disk swapping sounds really cool to implement but i'm sure it's not a simple as it sounds

[twilit smelt]

its simple once you understand it

[sterile frost]

many things are simple once you have a good understanding of them :p

sadly i lack the understanding lol

[twilit smelt]

it took me over a year to get it

a year and like 5 broken implementations

[coarse current]

I want(ed) to support swap in my OS

[sterile frost]

it took me like 3 years to write a shitty kernel that does nothing, crashes at the slightest inconvenience and i can't get anything done anymore because i set my standards way too high and idk where to learn about stuff

sooo

¯_(ツ)_/¯

it is what it is

at least im glad i can watch other people do cool stuff and maybe contribute

[warm pine]

things fall into place when you have the right design

[coarse current]

But got distracted (several times)

[twilit smelt]

the big thing about swapping is that it introduces shit tons of intricate little rules that you definitely did not follow and didnt even know you were breaking when you wrote your kernel initially

[sterile frost]

i remember the old keyronex that used to run bash and xorg and i think i can say i agree with that statement

[warm pine]

in my case that was gotten at by getting rid of the temptation to insert unwanted innovations and instead implementing something that hewed very closely to the mintia/nt vmm

[sterile frost]

it has progressed so much it's crazy

[twilit smelt]

this exact thing might be the ultimate source of the ke vs executive layering in NT

[sterile frost]

that makes sense actually

[warm pine]

it had the wrong design in a lot of places, but enough parts were right that it could get so far as that

[sterile frost]

is there a good way to learn about "the layering in NT" or did you just do your own thing in mintia?

[warm pine]

actually the VMM in it was neither fish nor fowl, i had studied the vms vmm tradition long enough by then that i did incorporate certain concepts like MDLs and the page refcount, but attached to a clock-style replacement

[twilit smelt]

now im gonna do a freaky UVM-NT hybrid bastard child where i use NTy policy (working set lists) and low level mechanism (page lists) but UVMy mid level mechanism (amaps and the fork impl)

[warm pine]

and incorporating those two concepts i finally learned the basis to implement page replacement and writeback without getting lost in crazy nonsense like extended page locks held during I/O (horrible!! shit!!)

[coarse current]

like there are probably so many considerations to do that with a microkernel

[warm pine]

a terrible beauty will be born

[twilit smelt]

i hold a per-file "paging IO lock" across file page writebacks but it only synchronizes against truncation, normal file reads/writes dont take that lock

they do take another lock shared (which truncation also takes exclusive)

[digital pivot]

iirc you supported memory compression or am i thinking of something else?

[twilit smelt]

but they dont take any locks that synchronize against page writeback

[warm pine]

it "worked" but would not work if the page was deallocated during compression

[twilit smelt]

pagefile writes do not take this lock bc that would cause deadlocks

and truncation is handled specially there

so that its not needed

[warm pine]

i think i do too, i forgot

[twilit smelt]

i mean its necessary

[warm pine]

or possibly i let the filesystem decide, there's something anyway

[twilit smelt]

i originally didnt have that and there was just a big race condition and i was re-reading "NT File System Internals" and it mentioned the paging IO ERESOURCE and what it was for and i was like ohhhh shit

i originally ignored it bc i didnt understand the difference with the normal file ERESOURCE (that is to say, rwlock)

and didnt notice the races between truncation and reading/writing (somehow)

i mean i think i synchronized it at a mechanical level in the fs driver so the on-disk structures were always sane but there was some data consistency race condition that could only be fully handled higher up

i dont remember what

[twilit smelt]

this book also gave us the cache manafler diagram

nt cache manafler is for sure its best component.

viewcache is a better name tho

whats a manafler anyway

[twilit smelt]

oh also

file page writeback also only takes it shared

ONLY truncation takes it exclusive

this could cause deadlocks bc theres paged code and data in the truncation codepath

if the modified page writer tried to take the paging io rwlock and a truncator had it

they might fault on something and block for pages and never get them bc the modified page writer is waiting for them to release the paging io rwlock

and this is why the modified page writer is split out into the file page writer

and it dispatches file page writes for that other guy to do

so that he can block on paging io rwlocks without blocking the guy who writes to the pagefile

this still has a potential issue which is if memory is 100% filled with file pages and the file page writer is your only hope for getting any

but this is impossible to happen in practice

kind of accidentally

just from the fact that theres not really any plausible sequence of actions to take that results in the system having only fluid file pages and zero fluid anon pages

youd inevitably fault in more anon pages on the way to trying to make that happen

specifically because there is paged kernel code and data that you cant stop from being touched if you take a page fault on a file page

so the paged kernel code and data actualyl serves an extremely subtle functional purpose lmfao

which is preventing this highly specific deadlock by stopping fluid memory from ever being 100% dirtied file pages

but thats not enough to rely on so you also should just throttle threads who create tons of dirtied file pages by forcing them to sleep for a few millseconds

i should add support for offlining a cpu to my scheduler

specifically so i can do the famous BeOS demo

https://youtu.be/BsVydyC8ZGQ?t=513

i have no other use for it

[coarse current]

isn't the interrupt handling a bigger problem with it than the scheduling?

[twilit smelt]

Probably

You'd probably need to message the drivers to get them to rearrange their own interrupts if they support such a thing

[coarse current]

Like scheduling is easy, just evacuate everything to a different CPU and stop it

But you have to have a really robust driver interface for interrupt renalancing

[twilit smelt]

Yeah I'll be able to do that too no problem

I just read about how windows does that exact thing the other day

It sends a plug n play STOP packet to any driver with resources taken out that need rebalancing, so in this case any driver that has an interrupt registered with the cpu being offlined, and the driver marks the packet completed after it quiesces and releases these things

It sends a START packet later (in this case presumably after the cpu is fully offlined) which causes the driver to re-request its resources, in this case getting some interrupts allocated on whatever cpu except the one that was offlined

And it has to restart it's in progress packets in a transparent way

So users don't notice

[coarse current]

I see ACPI in it

Yeah, after thinking about it it's not a big issue

[twilit smelt]

The STOP and START packets are very heavy handed tho

Drivers have to release literally every hardware resource they asked for (interrupts, dma channels, etc) without cancelling any ongoing user requests

And also save any state from the device that needs saving in ram bc the device might be getting powered off by user request or something

When it gets a START it has to transparently restart all of this

[coarse current]

it probably stems from the backwards compatibility (?)

[twilit smelt]

No

Well maybe

I was just gonna say it was cleanly designed like this relatively recently but by that I mean like 1998

[coarse current]

Was NT parking cores back then, or was it a convenient interface that they already had when that was implemented?

[twilit smelt]

I don't know

Also pnp was originally designed for win9x

[coarse current]

the famous bsod

[twilit smelt]

iiuc it was ported to NT after that

(Designed all along with the intent of moving it to NT later)

the original usage of the start and stop packets was for manually powering down a device at user request

it would send such packets to all device stacks downstream of a device being powered down

from the leaves of the device tree up serially

(but the leaves would receive these packets and process them in parallel as async IO requests, its just their parent wouldnt receive the packet until the children had fully completed it)

[coarse current]

Like when core parking was even added to Windows?

[twilit smelt]

no clue

[coarse current]

I'm suspecting around Windows 8?

[twilit smelt]

i intend to do it identically but more xnu-y and iokit-y

[coarse current]

(I have no idea how NT is internally)

[twilit smelt]

im gonna have a HUGE io system this time and it was already 15kloc last time w/o supporting all this plug n play junk and object oriented driver models and whatever so u know i mean business

nvm it was only 10k

largest file was IOPageFile.df at 1300 lines

Ke (basically the scheduler and exception handling) was 6900 lines last time and is 14,500 lines this time

So Far

if the ratio is the same for the IO system (it wont be) itll be 21k lines this time

ik ik youre not supposed to admire largeness

but largeness is cool when it reflects essential complexity

and not accidental complexity

terry "an idiot admires complexity. a genius admires simplicity" davis didnt understand this part bc of racism brain worms and a lack of appreciation for fred brooks

its funny he said that like templeos is a beacon of genius simplicity

despite that its still repeated by dummies about 100 times a day

usually to excuse being bad as just "no. its elegant. its simple"

"just like king terry would have wanted"

cuz they confused paring down accidental complexity with chopping items off their feature list

often not even frilly features either but like. "robustness"

cuz they have no concept of a distinction between complexity bc it does a lot, and complexity bc its badly written

a lot of ppl conflate the latter with the former explicitly and they do this under the like "well its just the unix KISS principle. do ONE thing and do it well. if youre doing lots of things then you simply have a bad program anyway"

somehow they dont realize the unix kernel at any point after the late 70s is itself a 100k, 200k, 500k, 1 million line behemoth that does a LOT

not to mention the toolchain its built with

the only value in KISS is the obvious "modularity" and "code reuse" stuff thats intuitive

using it as some iron law is extremely obselete thinking from like 1972 when they physically couldnt hold the results of each pass of a compiler in memory and had to write it out to disk and feed it into the next stage which was a separate program and so on

and nowadays is 90% of the time a cope by the mediocre to excuse sucking ass by invoking some like vague hero figures who coined a catchphrase in the 70s that they think is somehow supposed to still be automatically taken seriously today (prob cuz theyre too mediocre to use critical thinking)

honsetly the biggest reason i dislike microkernels is probably because they originated explicitly as an attempt to apply "KISS" to the unix kernel itself

which i think is like arbitrary dogma and completely technically inappropriate

and i think the best proof for that is literally every attempt to do a general purpose OS on top of a microkernel ever

no offense to the microkernelers in here

it reminds me of the "real communism has never been tried" thing

im sure this time there will be a good fast microkernel

[twilit smelt]

it occurs to me that if at some point im planning on doing linux userspace binary compatibility

i need to pick WHICH linux userspace

and then im doing distro dev

[warm pine]

it's the right way though

of course if you have a primitive dogmatic conception of what simplicity is then you paradoxically break the concept of simplicity

(maoists call this dogmato-revisionism, the inescapable tendency of dogmatism to also be revisionist, i.e. to deviate from the original concept while it pretends to uphold it. it's very dialectical but the same principle is also arrived at by very much not maoists, like popper with his paradox of tolerance. and in software you can see this a lot, for example how amigaos cultists try to claim the lack of virtual memory of amigaos as a feature, when in fact it was an undesired compromise made because adding an MMU wasn't practicable at the time. by reifying accident into essential dogma, they do a disservice to what amigaos was meant to be, what it could be)

and so the corollary to this is that you can't be truly faithful to a concept unless you are prepared to search for its fuller embodiment and application. take UNIX for instance

if you bear true fidelity to unix as an ideal then you have to support things like handles to processes being file descriptors. that's being more unix than historic unix itself

[warm mural]

I think a good microkernel is definitely feasible

for a general-purpose desktop OS, that is

just gotta know how to balance responsability between user processes and the kernel

[twilit smelt]

i have eliminated the big IO lock in xremu

idk why that was there for so long

it was a stopgap when it first became multithreaded and it basically just wrapped all the simulated devices

i just broke it up into per-device locks

on another note xremu probably needs like a total rewrite

it is so scummy

scrungly

inconsistent error handling (sometimes i handle malloc failure (at init time), other times i just abort()), inconsistent code style, generally ugly code

[sterile frost]

RIIJ

[twilit smelt]

it seems to have had a macroscopic effect

when i removed it, some text that ive never seen mixed together (bc of race conditions; multiple cpus printing simultaneously) was mixed together consistently

suggesting increased parallelism

also finally motivated me to put hooks around my RtlPrint function that let the client (in this case the kernel) take a lock to guard the line if it wants

i assumed it wouldnt have had much of an effect bc accessing devices is pretty infrequent and most of the time each cpu lives out of its own cache (which is locked per-cache-set)

but ig it did

[sterile frost]

wow i had a hunch the other day :^)

[twilit smelt]

it didnt help with the fireworks jitter

[sterile frost]

ah :(

but improving the parallelism at least a bit is still a good outcome i guess

[twilit smelt]

but thats on my 11 yr old intel imac, on my m1 macbook i can barely notice the jitter

so im p sure its just a result of a really shitty ancient intel cpu from 11 yrs ago not being good at running the cpu threads in a physically simultaneous manner and so they get stuck on spinlocks and IPIs for a long time, waiting for the other guy to get scheduled in on the host

and the pause trick (which causes the cpu thread to sched_yield() if you execute like >256 of them within a millisecond, and i execute this on each iteration of a spin that waits for another cpu) only goes so far

it does go pretty far though

pre-pause if i ran the fireworks test with 8 cpus on my intel imac it would BARELY even run

now it runs and is only slightly more jittery than 2 cpus

[sterile frost]

how many cores does the imac have?

[twilit smelt]

as a direct result of the pause thing

4 lol

[sterile frost]

4 physical 8 logical

or 2c4t

im assuming former

[twilit smelt]

former

[sterile frost]

makes sense

maybe kernel overhead

but i dont think it would account for that much

[twilit smelt]

it might. im running a version thats like 5 years newer than the last officially supported OS on this machine

[sterile frost]

i mean at least one of the logical cores is shared between the "guest" and the rest of the os

probably two or three though

but im also assuming they aren't being utilised 100% by the emulator

[twilit smelt]

basically im on the newest macos version on a 2014 intel imac

this is not very performant

[sterile frost]

yeah that sounds less than optimal

ever thought about just using linux on it?

[twilit smelt]

i only updated (which was only possible with some third party patching software) bc i was worried about malware

i meant to imply i want my experience to improve

not get worse

my bad

[twilit smelt]

also the benefits will be way more visible when im doing fun async IO stress tests and stuff

as of some time ago all disk IO in the emulator is async by enqueuing an SDL timer with a timeout of 0 which effectively causes it to get instantly dispatched on SDL's timer thread

so the cpu threads dont have to take the burden of the read() and write() syscalls and stuff

previously itd be highly likely for this work thread to contend on the big io mutex with a cpu thread (rendering a lot of the benefits null) but now its much less likely

cuz it just takes a disk controller mutex instead

[mortal thunder]

finally

i remember mentioning this to you a few mo ago

why dont you just add a stringify type preprocessor feature

like in C, #paramnamehere turns whatever tokens you passed into it into a string

[twilit smelt]

bc i evidently dont need one

[mortal thunder]

evidently you do need one that strcat ( strcat ( ) ) looks ugly

[twilit smelt]

u can do it the way u like it in ur own compiler

[mortal thunder]

of course

[twilit smelt]

which is a good response to annoying aesthetic nitpicks

but if u ever see me say something like that in response to some major functional deficiency i want you to shoot me

bc it means ive died

and something evil took my place

this isnt that tho cuz it works just fine

you just find the way it works ugly

which idc about

i listened to ur functional critique about the asserts needing to exist but this aesthetic one idc about

u see how that works

i should probably add the function name to the assert message

[mortal thunder]

makes sense

[signal zinc]

i still don't get how this works 🥺 stir cat

[mortal thunder]

it st(i)rs two cats together into one big cat

[twilit smelt]

bom

just the file name and line number were too ambiguous

bc that can change from version to version

[twilit smelt]

the preprocessor directives have a lispy syntax

( STRCAT a b ) yields a string ab

so this is basically doing like

EXPR_STRING = "\"" + expression + "\""

[signal zinc]

awawa

[craggy spire]

Windows 7

It didn't rebalance the interrupts back then

The main idea is to disable thread scheduling on the parked core so that it can potentially idle

[mortal thunder]

all of them

and by that i mean it shouldn't matter if you just emulate all the syscalls

[mortal thunder]

but why are you using such an old imac if you have the macbook

[twilit smelt]

Big 5k monitor

[mortal thunder]

when in 2014 was it released

https://discussions.apple.com/thread/255471650?sortBy=rank

you might have considered this but just wondering

but uh it seems like there are some seemingly arbitrary restrictions on what versions actually work

like "you can only use target display mode if the source is running macos catalina or earlier and the mac was introduced 2019 or earlier"

[twilit smelt]

I just remembered realloc exists

Leaving this as a note here for when I have time to go and use it in appropriate places

Consequences of being dumb

[white valley]

what years of linked list usage does to a kernel dev

[mortal thunder]

wdym

like your design didnt account for it?

[twilit smelt]

I just didn't implement it and had been doing it the hard way in a few places

Just a dumb thing

[white valley]

the hard way as in like

classic malloc memcpy free

[twilit smelt]

Yeah

[white valley]

wouldnt it be like more efficient to keep the physical pages

and only realloc the virt pages

[twilit smelt]

For large alligations yeah

The ones that were made page aligned to begin with

For smaller ones I'll just check for enough free space to the right of the block and try to extend it into there

If not then I'll just fall back to a malloc memcpy free

[twilit smelt]

seeing this feels like painstakingly climbing down a cliffside of the grand canyon and then up the other side and then looking back across and theres a tech bro going "guys. its HIGH SIGNAL to be on the other side of this canyon. all you have to do is jump across" and then 685 people jump to their deaths

[warm pine]

it seems to be a universal rule of osdev that any tutorial will cover legacy x86 minutia in great depth (but often not very accurately) then quickly peter out before it gets into actual osdev

just realised mintia is one letter away from minutia

lmao

aarch64 by far the toughest target i have targeted and i've done 4 targets

[twilit smelt]

i think about that every time i write the word mintia to the point where ive considered calling the internals book "MINTIA Minutia"

[warm pine]

you think idt and gdt are hard? wait until you see aarch64 cache modes and maintenance and shareability and so on

[warm pine]

it's too good to pass up, beats mintia internals by a country mile

[twilit smelt]

these people think theyre smart bc they push out slop (usually vaguely right wing coded) to a crowd of useful idiot cs majors who lap it all up

ppl rly similar to that poster are currently taking control of the apparatus of the federal government

so thats cool

[warm pine]

it's just a bunch of empty words they're repeating because they've heard them said before by others - nothing but miming

[twilit smelt]

dumb silicon valley billionaires dreamt about libertarian technocracy and when they tried to make it real the best they could do was authoritarian tech-bro-cracy

[warm pine]

herbert couldn't let a reference to x86 go by without having to pipe in with trite commentary

[warm pine]

i think they talked about this in The Californian Ideology

[twilit smelt]

the reason its scary is that theyre doomed to crash and burn because they arent as smart as they think they are and reality will eventually catch up to them. they forgot that theyre just marketers and accidentally put themselves in positions where they now have to be technically competent in order to succeed

this would be amusing if it werent now tied up with the fate of the western world

smth interesting is if im wrapped in a blanket and i throw it back onto my bed my headphones gently play static for a few seconds

wonder what static electricity bs is going on to cause that

[warm pine]

musk will get a rude awakening with doge

he is a bit like the archetype of the manager who never spent any time on the shop floor so he thinks he can cut this or that and change whatever and it'll all improve things because he's smart

but the depth of arrogance is much, much worse because of the tech background

[twilit smelt]

he's worried about assassination but seems to think the most likely culprits would be edgy posters on twitter and not the CIA

i dont think he knows what game he's gotten himself into trying to use his financial and soft power to reshape the federal govt in a way that could risk america's national security

or he's mistaken the CIA's bureaucracy (which probably feels mostly transparent to the authority of the executive) with its actual actors

they could be drafting up contingency plans to violently counter-coup DOGE in an office building that the bureaucratic administration of the CIA doesnt even know exists

by design, bc the executive changes every 4-8 years and thats far too fickle and risky to rest the national security of the country on, the only real solution (which ppl would have realized over a century ago) would be to have layers of secrecy such that the president cant know everything thats going on no matter how much he'd like to, and what he doesnt know about he cant control

i mean this is intelligence agency 101 shit that the CIA surely implemented decades ago at least

elected civilian administration just cant have that much control or knowledge over these things - it just wouldnt work well for very long if at all

a really robust security state has to be inherently undemocratic in some way probably

robust secrecy is undemocratic

inherently

is what im saying

its also required to maintain the long term security of a global hegemon

[twilit smelt]

so theres a high chance this is happening somewhere

it probably varies over time how much the "director of the CIA" knows or has control over what the CIA is doing

at a time like this its probably at an all time low bc the people would adjust around their perceived trustworthiness which would not be high at this moment, being a cabinet pick by an administration that is seemingly actively working against our national interests for some reason

it always struck me that "the director of the CIA" just cannot know that much about what its doing. like. ironically a fully-informed central administration is antithetical to the whole purpose of its existence

like you probably have lower to mid level administrators who have been there for decades and have the trust of "their people" and so they probably know and can direct certain things

but some bullshit cabinet pick by an idiot president probably has substantially less influence and might be judged very low on the "need to know" scale

i bet the FBI is a lot worse at keeping secrets from itself though

it probably does have a fully informed central bureaucratic admin and thats how trump was able to ask them for a list of agents who worked against him and get a complete answer

for the CIA something like "tell me all the agents who work on sensitive intelligence about Russia" is probably more of an unanswerable question even if they wanted to

if you can even answer a question like that youre probably a shit intelligence agency

maybe someone like the modern Mossad outclass the modern CIA at that kind of secrecy though, i really would not know (and by definition maybe nobody knows.)

it would explain a lot if that were the case

honestly if literally everything going on in the US right now were the result of an extremely successful Mossad operation to turn it into a gigantic client state of Israel i would only be like 70% surprised

they do stand to gain the most by far

ppl like to point the finger at putin bc he gets to keep some territories of ukraine and that is a pretty substantial gain which makes him a good suspect

but israel getting our unconditional direct diplomatic and military support with literally anything they do ever is probably a relatively much larger one

i dunno. its difficult to explain whats happening, i dont want to think that the silicon valley "technolibertarians" are winning this hard just by their own hand bc theyre so stupid its painful to imagine

imagining them as useful idiots who are being exploited by a foreign power and tricked into thinking its all their own idea (which would be easy, because they are stupid) makes more sense

[white valley]

what do you think about trump saying fuck u to the court and deporting people who havent even be proved to be gang members anyways

and trying to deport someone with a green card cuz they protested against israel

[twilit smelt]

i mean those are transparently authoritarian moves

and a level of deference to a genocidal foreign power that is absolutely insane and probably unprecedented in american history

and difficult to explain if there isnt some kind of grasp on this administration

[white valley]

do you think this'll be used to deport political rivals

[twilit smelt]

no those will just be made irrelevant

the democratic party is literally too impotent to be a threat so the old assholes will probably just retire and live out their lives peacefully in the first years of the trump dynasty or something

there are no politicians who can stop these people

[white valley]

lmao it's literally the same here, the guys who're supposed to show the most opposition rn dont even treat the protests like protests

[white valley]

do you have laws saying a president cant run for like more than 2 terms

[warm pine]

they claimed biden was an old man taken advantage of by others in his government

well, Mary Anne's boy, he seems to be that to a tee

[white valley]

cuz we had those here and mr dictator didnt care

[twilit smelt]

a constitutional amendment says that yeah

but the constitution has been openly decried as obsolete by people who now hold a shit ton of federal power

the aforementioned silicon valley idiots who kind of snuck in the back door behind trump

led (publicly) by musk (thiel might be more influential behind the scenes)

and mainstream republican politicians have called for trump to run for a third term

[white valley]

how do you think this'll stop

[twilit smelt]

to unanimous cheers from their constituents

one of them was drafting a constitutional amendment to make the 2 term limit a 2 consecutive term limit

so you can have another 2 terms if theres a gap inbetween lol

the extra time would be useful for further dismantling the federal government and reprogramming half of the country into thinking federal democracy is actually woke and anti-freedom

and trying to set up a new form of govt with trump as CEO and musk or thiel heading up a board of directors or something like that

thi sis obviously stupid and i dont expect it to work out how they plan but i fully believe the silicon valley billionaires want to do exactly this

but like i said theyre probably being puppeteered (potentially without even knowing it) by the intelligence services of a foreign power like israel or russia or both

and the real goal is just to convert america into a rabidly conservative and openly imperial authoritarian country that will never stop serving the interests of (israel/russia/whatever) for the foreseeable future; the point of the authoritarianism is just to give a certain segment of the elite class (carefully installed) more power to serve your interests for longer

which is probably the best plan if your (potential) enemy is a literally unbeatable global hegemon

just turn them into ur bitch while tricking them into thinking theyre just awakening a newfound national pride

now youre the unbeatable global hegemon for like 1 millionth the work

for a while at least

[white valley]

i see

[white valley]

and?

[twilit smelt]

i think whatevers happening will reach a conclusion with minimal interference from opponents

[white valley]

no impeachment?

[twilit smelt]

uhhh no

not even close

lol

[white valley]

damn

do you think luigi'ing trump/musk would do anything

or would they be instantly replaced by worse people

[twilit smelt]

probably not

i think if you went that route youd have to push a red button that would immediately disappear like about 50 people simultaneously

this isnt possible so.

[white valley]

the 3 letter organizations would never

[twilit smelt]

literally the (actual) deep state is the only hope rn lol

the only potentially non completely impotent opponents of the actual people responsible for whats happening

whoever they are

assuming its not just straightforwardly the dumbass silicon valley billionaires which again i dont want to believe bc its so stupid

i hope it is bc thats a MUCH more beatable opponent

DOGE vs CIA: FIGHT

[white valley]

also isnt CIA like not supposed to be fiddling with internal affairs

[twilit smelt]

yeah but they always have lol

[white valley]

why not this time

[twilit smelt]

they might be

it doesnt make any sense for them to be responsible for whats going on bc it only diminishes their own power and prestige

which is why im assuming its a foreign power

[white valley]

how are you so sure theyre not people with MAGA hats sitting around a cybertruck and worshipping it

[twilit smelt]

bc modern CIA agents have the same make-up as like modern math and (actual) programming geniuses

that is to say theyre like 40% trans girls

and are extremely liberal

[white valley]

how do you know

[twilit smelt]

bc some internal CIA chats leaked recently (by the trump administration, as culture war fuel) where it was literally 100% trans girls talking about gender affirming surgery and stuff LOL

in perfect discord speak

right-wing media outlets like Fox News and The Daily Wire picked up on Rufo's story — as did unelected Trump administration official Elon Musk, who suggested the revelations show that "MAJOR housecleaning is needed" at America's spy agencies.

[white valley]

muh, this is proof that trans people were invented by the CIA!!

when they say "housecleaning" do they mean filling all agencies with only cishet white guys wearing maga hats

[twilit smelt]

Another intelligence official boasted that genital surgery allowed him “to wear leggings or bikinis without having to wear a gaff under it.”

These employees discussed hair removal, estrogen injections, and the experience of sexual pleasure post-castration. “[G]etting my butthole zapped by a laser was . . . shocking,” said one transgender-identifying intel employee who spent thousands on hair removal. “Look, I just enjoy helping other people experience boobs,” said another about estrogen treatments. “[O]ne of the weirdest things that gives me euphoria is when i pee, i don’t have to push anything down to make sure it aims right,” a Defense Intelligence Agency employee added.

agents from several agencies were involved in these chats so its not just the CIA lmfao

the fact it was an inter-agency chat is probably how it was unsecret enough for ppl as untrustworthy as the trump administration to be able to get a peek at it at all

[twilit smelt]

unfortunately this quote is from a conservative news outlet who go out of their way to misgender the agents

its an accurate description otherwise

[white valley]

yeah i noticed that

[twilit smelt]

5 transgender cia agents are assembling poison darts in a secret warehouse rn

[white valley]

first transgender us president when

would be a bigger deal than obama

[twilit smelt]

this whole thing might come down to a secret intelligence war like a dumb spy movie

if its just against DOGE theres nothing to worry about bc those guys are idiots who dont understand the concept of opsec and half their girlfriends are probably spies already

FSB or Mossad are scarier as i already said

netanyahu probably wasnt even briefed on the pager thing until an hour before they set it off

the guy who briefed him probably wasnt briefed until an hour before that

i do understand opsec which is why me expressing my views on all of this openly is proof im a lazy asshole and a complete non-threat. i am not worth ur time scary spy people.

[white valley]

EXACTLY what a mossad agent would say

[twilit smelt]

my understanding is that they have blanket approval and a budget to develop things like this and they only need to tell their bosses in order to get final approval to pull the trigger so to speak

actually this is absolutely necessary or nothing would ever work, as soon as information reaches the elected administration of any democratic nation the odds of it inevitably leaking accidentally or otherwise probably goes up to 100%

obama probably didnt know they knew where bin laden was until they asked for approval to go kill him

well i mean ig that ones obvious

he def knew they were looking for him lol

only cuz everybody did though

[tiny swallow]

at least you got a bureaucracy

[mortal thunder]

kinda already happening, students with green cards for example are being deported for protesting in favor of palestine

[mortal thunder]

I don't think it will

[white valley]

u dont even live there

i asked will cuz he has the most exposure to us politics

[mortal thunder]

no but i live close to russia which the US is a de facto ally of

The russian regime wasn't taken down in 35 years

what makes anyone think musk's regime will be

even if trump dies musk will probably still have the same amount of power with jd vance as president

[coarse current]

i think you're overestimating russia

[green pelican]

im certainly not on isreals side in this but you have to be delusional if you think you can immigrate to a foreign country and then start protesting against them and their allies without repercussions

imagine immigrating to china and then going to pro taiwan protests, you'd be put in jail forever

have some common sense

[mortal thunder]

if it's not against the law then it should be fine in my opinion

and of course you dont overexaggerate

[green pelican]

the constitution doesnt apply to people who arent citizens

this is the same in almost every country

[twilit smelt]

Yes it does

The legal principle is that if someone is subject to our laws then they also are guaranteed the rights that restrict those laws

Bc they are part of the law!

And you're not supposed to be able to pick and choose what parts of the law you want to apply to someone

[twilit smelt]

China isn't the standard we should be holding ourselves to lol

[twilit smelt]

Also the bill of rights is meant to legally enshrine natural rights that belong to all humans by virtue of them being born human

[green pelican]

yeah thats fair, my point was more that being deported is the nicer end of the spectrum

[twilit smelt]

You should be able to come to America and peacefully protest their support of a genocidal regime lol

Bc it's a free country. We always say "come over and have some freedom!!" and it's lame to arbitrarily go "oh nvm no freedom for you you criticized our evil ally"

[green pelican]

america is not known to be the most consistent country ngl

im not arguing if its right or wrong, im saying that its kinda insane to expect nothing to happen when people do these things

[twilit smelt]

No it isn't

Please stop talking before I stop liking you

This is a gradual erosion of rights issue, because in America no there's not really any precedent for deporting legal residents for protesting, until now

What you're saying is Russia or China logic

[twilit smelt]

Also it is very likely unconstitutional bc the first amendment applies to any enforcement action the government takes and the question is whether this counts as such. The answer is probably a slam dunk "yes" considering the relevant governing entities said OUT LOUD that it was specifically a punishment for peacefully protesting (ie, exercising free speech rights) lol

They probably don't even want it to succeed they just want to put on a show for their moronic base

[heady bobcat]

didn't Putin do this?

[twilit smelt]

Yeah I think so

[raven drift]

The chances for trump to amend the constitution to allow such a thing are basically zero though

[green pelican]

my favourite conspiracy to come out of twitter is that trump shouldn't have been allowed to run in the 2024 elections because he won the 2020 ones

[mortal thunder]

^^^

[mortal thunder]

I wouldn't be so sure

Not mentioning that he's likely going to try

[raven drift]

Amendments need a 2/3 majority

[mortal thunder]

The SCOTUS is pretty much in his hand

Isn't it?

[green pelican]

certainly not a 2/3rd majority of it

and i doubt many republicans would vote for it anyway, the constitution is like the one thing you dont touch

and on top of that voting for that kind of thing is political career suicide, you'd never be elected by your state again

[heady bobcat]

In the Swedish constitution it is written that an amendment needs to pass the vote two times, with an election in-between (obviously in legal-speek and not as straightforward as this, though). There's also a process where proposed changes to the law are sent for review to interested institutions and governmental bodies months before they are voted on. Is this just not a thing elsewhere?

[green pelican]

iirc in canada its a 2/3rd majority vote as well, but again its career suicide even if it passes. everytime the constitution has been amended the person that did it was shunned out of politics pretty quickly afterwards

[heady bobcat]

if the orange man tells them, it is effectively career suicide, at least for the forseeable future, not to.

[green pelican]

whether thats enough now that so many politicians have piles of money and can just leave to live elsewhere is another question

[cunning mortar]

yeah lol

Apache is obviously showing he has no fucking idea what he's talking about

also fuck Israel, free Palestine 🇵🇸

[heady bobcat]

in a sensible world, what he's saying is true, though.

[cunning mortar]

yeah well

he's talking about American politics in 2025

[green pelican]

i sometimes forget that most of the world is in a terminal state of delirium

my bad

[cunning mortar]

hell, not even just Republicans

Democrats will vote for what orange man says too

because bags of cash

[green pelican]

forget age limits on politicians, they should enforce that everyone gets 9 hours of good sleep before they're allowed to vote on anything

[twilit smelt]

20 years ago you'd be right but rn Trump is god-appointed emperor of the republicans

[cunning mortar]

though i doubt he's going to ever get a constitutional amendment through anyways

but that doesn't matter

Trump has been ignoring the constitution

[green pelican]

it would be nice if he could stop putting tarrifs on random things

[twilit smelt]

Amazon made me pay a tariff to buy my Canadian friend some Korean skincare goop :(

[green pelican]

damn that was quick on amazons part

[cunning mortar]

if you're American support AOC and Bernie

[twilit smelt]

I think ppl are going to just keep voting for moderate do-nothings who get obliterated by trump and friends until it's too late

[green pelican]

vote for howard dean

i have no clue what his policies were but i have a couple of tshirts and they've been very good quality

[cunning mortar]

dissatisfaction of Democrats with the party is at an all time high, if there is a chance for people who are not moderate do-nothings to shine it'd be now

so support them while you can

isn't it true that the brightest of stars shine in the darkest of nights :^)

[white valley]

nope

[green pelican]

there are solutions to many political problems but discord TOS prevents me from talking about them

[cunning mortar]

you're Russian you probably didn't even get the reference

[white valley]

1. kill everyone i dont like
2. kill everyone i dont like
3. kill everyone i dont like

is that your list

[green pelican]

no those are solutions to personal problems

[twilit smelt]

I'm not even convinced that there will be a next presidential election in any meaningful way

[cunning mortar]

well the mid-terms are first

we'll see with those

[white valley]

yes im the most russian person here

[cunning mortar]

sorry, you're Turkish

same thing

[green pelican]

america hasnt had an election since 1960

[twilit smelt]

Bc capital is 100% on board with god emperor trump and the most powerful forces in this country will be trying their best to suspend or subvert elections while their only opponents are impotent Democratic Party leaders

[white valley]

yup completely the same thing

[green pelican]

hasan piker country

[cunning mortar]

yes sure, we can be analytical smartasses about it, in which case you'd be right, but with that attitude all you got's a self fulfilling prophecy

[twilit smelt]

We will probably increasingly see the tech billionaires using their centralized control over the entire internet toward this purpose

They already are but I mean even moreso

[green pelican]

i was very suprised to see how quickly tech giants turned coat when trump got elected

google renaming the gulf of mexico to the gulf of america in like a week was not something i thought would happen

[twilit smelt]

It's because they are all "technolibertarians" who think they can co opt the Republican Party to destroy the government and have their ideology rise from the ashes

[green pelican]

i get that these megacorps are all about making money but i would've thought the amount of internal pushback would've slowed stuff down alot

[twilit smelt]

No there's just been a weird political extremism brewing in Silicon Valley circles that kind of snuck up on everyone

They've been saying it out loud for decades but ppl brushed it off as "haha what loony eccentric billionaires how quaint" until extremely recently it got less cute when they started concentrating their power on implementing it

At some point they recruited Trump and that gave them the opportunity to do all this

In actuality they don't like either the conservative base who they think are mouth breathing tribalistic cavemen nor the liberal base who they think are deranged woke communists

They like rich people (themselves)

Who they think are superior humans

They think wealth is the highest measure of merit and that they should get to rule the modern technological world they think they built, for everyone's good

[twilit smelt]

The conservative base is easier to manipulate due to being stupider which is why the tech guys are teamed up with the republicans now

You could see a demonstration of this division recently with the like H1B visa controversy lol

The tech billionaire part of maga was like "we love Indian immigrants bc we can pay them less and they work harder" and the racist part (the entire rest of it) was like "what????? no! we hate Indian immigrants bc [insert racist tirade]"

[green pelican]

incredible how you just cant win in america

[white valley]

it's a big club

and you aint in it

[cunning mortar]

https://www.youtube.com/watch?v=NZSN1H0Jt-Y

[green pelican]

at least in eastern europe bribing the police and politicians isnt reserved for the super wealthy

[white valley]

lmao

[twilit smelt]

The reaction of most liberals to Bernie is still "what a crazy old man, if it wasn't for him Hillary might have gotten more votes >:("

I think the Democratic Party will just disintegrate and become an irrelevant husk and it'll take like another 15-20 years for a new leftist party to appear that has any juice

I hate to say it

Average liberals are going to keep smugly accusing ppl further left of sabotaging them and keep trying to pander to an imaginary moderate base that no longer exists and this will continue until they're outnumbered within the left by young ppl who have been more radicalized by what's coming

[cunning mortar]

the analysis is correct

i hope it happens faster than you think it will

[twilit smelt]

And that itself might be pretty ugly too

No moderates left on either side means appeasement stops and things maybe get violent

Might be a civil war by the time this saga comes to a close

[green pelican]

born too late to explore the world, too early to explore the galaxy, but just in time to watch america balkanize

[twilit smelt]

I think if it happens it could still be quite a ways off and trump would be long dead by then

[green pelican]

although i doubt it ever will, too much important stuff depends on america staying mostly in one piece

nothing ever happens and all that

[white valley]

btw how things are going in your country is very similar to turkey

minus the tech bro in government part

[twilit smelt]

There's also a slim chance trump dies of old age and the right loses all its juice and things return to status quo but if there was ever a window of time where that'd happen it's probably rapidly closing

[white valley]

but the "fascist guy takes over and fucks up the country while the leftist party is too incompetent to do anything significant" is pretty similar

[twilit smelt]

Permanent cultural and structural changes are ongoing that are ruling this out or already have

The last straw might have been capital throwing their lot in with him

That might be the point of no return

Bc that's a very powerful force that'll work to continue what he was doing after he's gone

Also musk has an extra motive aside from "rich people are ubermensch who deserve to divide the earth amongst themselves and rule as dictator CEOs of corporatocracies"

He probably convinced himself that he needs to control the govt in order to direct its resources toward like mars colonization which he thinks is humanity's only hope

[mortal thunder]

what does it matter

your job is done

[green pelican]

most politicians are in it for the power and influence, if you're not elected again you lose those things

[mortal thunder]

the solution is simple

modify the constitution to allow the president to personally appoint [insert position here] without Congress or a vote

you lose popularity but if it passes no one can get rid of you without violence

honestly i kinda wanted to start fixing shit in my life but recent events and russian threats have completely killed my motivation to do anything

if im going to die in 5 years (please dont) then why should i bother doing anything other than the bare minimum

[cunning mortar]

damn holy shit what attitudes i see

[mortal thunder]

now i hope that russia DOESN'T invade but yknow

[cunning mortar]

no wonder how Trump is in power

[green pelican]

because your ancestors didnt survive wars and famines just for you to up and kill yourself what kind of question is that

2000 years of people struggling for a better life for their kids just for some idiot to go "oh no the world is bad time to become a genetic dead end"

[white valley]

they didnt say theyre gonna kill themselves

they just asked why they should be doing anything with their life if theyre going to die in a short time anyways

[green pelican]

doing nothing and letting yourself be trampled by circumstance may as well be equivalent

[mortal thunder]

i'm not american

[white valley]

were you romanian

[mortal thunder]

i am

[white valley]

will you be romanian

[mortal thunder]

i am already what do you mean

[white valley]

gotta make sure you're not one of them greeks

[green pelican]

lmao

[mortal thunder]

true

i don't want to get enrolled in the romanian army, but if i have no choice

then i will try

[white valley]

also if russia doesnt invade then you're still not safe

[mortal thunder]

about as safe as now

[white valley]

one day you might trip going down the stairs and land perfectly on the tip of an unnecessarily sharp umbrella

[green pelican]

if you get conscripted tell the constription officer you'll shoot your commander the second you're handed a gun lmao

they cant really take you after that

[mortal thunder]

hell no

[white valley]

and spend years in jail?

lmao

[coarse current]

it doesn't work like that

[mortal thunder]

they will just take me to jail

also i would never

[green pelican]

out of the 500,000 people that dodged the vietnam war draft only 3000 served jail time

its an empty threat

[coarse current]

but like also at that point you're dealing with armed forces

[mortal thunder]

did all 500k use that threat or what

no

[white valley]

uuh

thats not how things work in the balkans

unfortunately..

[green pelican]

die for a country that hates you or spend time in jail is a pretty easy choice either way ngl

[coarse current]

the best is to try and avoid both

[teal trench]

i wonder how other inmates treat people who are in jail for dodging mandatory service

[green pelican]

well yeah ideally flee to another country long before the draft becomes a thing to worry about

[white valley]

uuh just dont tell em you're in for dodging mandatory service??

[mortal thunder]

pretty badly id assume

[mortal thunder]

theyll think you did worse

[white valley]

tell them you robbed a bank or something

[white valley]

anyways worst case they declare you a traitor and beat you up

depends on how nationalistic they are ig

[mortal thunder]

this too

[white valley]

anyways i doubt theyll have enough space to jail everyone who dodged their drafts

and you should easily be able to like cross a border illegally

[green pelican]

romania is part of the EU isnt it?

[raven drift]

Stop doom scrolling and watching news

[coarse current]

yeah, you can't flee to other countries in eu

[white valley]

idk but my friend crossed the border to bulgaria with an out of date passport and many liters of wine

[earnest zenith]

that's not good

look

[raven drift]

The world is still a much better place than 10/20/30/40/... years ago

[earnest zenith]

your great grandparents lived through way more cataclysmic events than the stuff now

[white valley]

you could flee to turkey

[green pelican]

theres like 3000km of land border, there are plenty of ways to get out

[earnest zenith]

imagine living through WWI and WWII in Europe

[cunning mortar]

yeah i know, but you're sort of representative of what the average young American may feel

[earnest zenith]

and honestly, the only option we all have in this regard

or the most productive option in any scenario

is to carry on with life

[mortal thunder]

that's true

i never implied i would kill myself

[earnest zenith]

oh

[mortal thunder]

just that i might die by russian

[earnest zenith]

not in that way

I mean

[mortal thunder]

either missile or gun

or mine

[earnest zenith]

carry on with day-to-day life

and on improving yourself personally

[green pelican]

hey did you know one of the main things the CIA learned from their experiments with mk-ultra is that people who are frightened and who feel helpless are easier to control and manipulate

fun fact

[mortal thunder]

wow

[mortal thunder]

possibly

[white valley]

in any case it doesnt help to be like "oh no i might die!!"

[mortal thunder]

and i should think

[white valley]

just live your life to the fullest

and you can always flee to turkey

[mortal thunder]

maybe russia wont invade

after all they havent for 3 years

[white valley]

we dont care too much about illegal immigrants (by we i mean the government. they love em)

[green pelican]

i dont think russia has another war in them for the forseeable future once ukraine is done

they've cratered their demographics

[coarse current]

tbh I have no idea what's going on at this point

with russia and neighbouring countries

[earnest zenith]

I doubt that
the Democrats have been in a way worse position and in way more disarray than nowadays
there's been 180 years of events

[green pelican]

they'll probably win parts of ukraine but they're running out of people to throw into the meat grinder

[white valley]

think of 2016

you're playing agar.io with a buncha friends

and theres a huge fucking guy on the map

who ate you in the past

[coarse current]

i am russian

[white valley]

and now he's not interested in eating you

but you dont really know when he might turn

[coarse current]

but yeah russia seems to have big demographic and economic problems from what i hear

like the situation is bad

i get surprised when i hear about "russia interfering in elections" and stuff like that, the country is just not that powerful in my eyes

[green pelican]

below the age of 30 women outnumber men by like what kind of ratio

its something insane like 65% female in that age band

[white valley]

i'd get my eyes checked if i were you

[coarse current]

it's not ww2 territory

ok maybe bad but not terrible

[green pelican]

just a bit worse than canada where 60% of the population under 30 is male because of the amount of immigration from india

[white valley]

i really dont understand countries taking immigrants from places where the culture is like the exact opposite

[green pelican]

its great how many countries are seemingly trying to speedrun total demographic collapse