Ultra | Operating System Development | Page 16

pearl meteor May 18, 2026, 3:25 PM

#

yes

sinful river May 18, 2026, 3:25 PM

#

2 levels with 2 bits:

root bitmap: 01
[0] sublevel bitmap: 10 (represents two CPUs, one has quiesced the other hasn't)
[1] sublevel bitmap: 11

When the other cpu in [0] notices it was the last one to set its bit it'll set the bit in the root bitmap

prime wraith May 18, 2026, 3:25 PM

#

I see

sinful river May 18, 2026, 3:25 PM

#

yeah they configure fanout based on NR_CPUS

pearl meteor May 18, 2026, 3:25 PM

#

yea

prime wraith May 18, 2026, 3:25 PM

#

sinful river yeah they configure fanout based on NR_CPUS

Its probably max online cpus or smth

#

Because nr_cpus is like 8k

sinful river May 18, 2026, 3:26 PM

#

no it is NR_CPUS

#

it's compile time

pearl meteor May 18, 2026, 3:26 PM

#

yes

prime wraith May 18, 2026, 3:26 PM

#

Oh

sinful river May 18, 2026, 3:26 PM

#

last i checked at least

pearl meteor May 18, 2026, 3:26 PM

#

but obviously not all structs are allocated

sinful river May 18, 2026, 3:26 PM

#

well even if they were it wouldnt be that bad

#

even with 8k cpus

prime wraith May 18, 2026, 3:26 PM

#

I just dont see the point since u know at boot time how many max cpus u can have

sinful river May 18, 2026, 3:27 PM

#

sinful river 2 levels with 2 bits: ``` root bitmap: 01 [0] sublevel bitmap: 10 (represents t...

this is a really shitty explanation but like now you can see that only two cpus have to contend on the global bitmap instead of 4

#

tho in practice I think the bits are all 1s and the cpus clear the bits

pearl meteor May 18, 2026, 3:27 PM

#

prime wraith I just dont see the point since u know at boot time how many max cpus u can have

the tree is fixed

prime wraith May 18, 2026, 3:27 PM

#

sinful river this is a really shitty explanation but like now you can see that only two cpus ...

Yeah I see

pearl meteor May 18, 2026, 3:28 PM

#

sinful river tho in practice I think the bits are all 1s and the cpus clear the bits

yes

#

because its easier to test for 0 than for a specific amount of 1s

sinful river May 18, 2026, 3:29 PM

#

well yeah but you can also do ~0

#

if you fix the bitmaps at u64s

#

which I think they do?

pearl meteor May 18, 2026, 3:29 PM

#

what if you have 4 cpus

sinful river May 18, 2026, 3:29 PM

#

ah right

pearl meteor May 18, 2026, 3:29 PM

#

and the bitmap is 64 bits

sinful river May 18, 2026, 3:30 PM

#

idk if I'll implement rcu in my current project, im thinking the alternatives look nicer

prime wraith May 18, 2026, 3:30 PM

#

Which

sinful river May 18, 2026, 3:31 PM

#

freebsd GUS comes to mind

#

and xnu's SMR which is basically the same thing

prime wraith May 18, 2026, 3:31 PM

#

Ah

pearl meteor May 18, 2026, 3:39 PM

#

and the way grace periods are started it's also a little bit insane
so, in each rcu_node and rcu_data, there are two fields, gp_seq and gp_seq_needed
gp_seq is the current grace period, and it's modified by the rcu gp kthread. the lower 2 bits, if non-zero indicate that a grace period is in process (aka. the rcu gp kthread is executing, and waiting for all cpus to report quiescence). a cpu that wants a start a grace period, samples without taking any locks and using read_once the gp_seq of the root, does a simple addition in order to get the number for the next grace period, and updates the gp_seq_needed from its rcu_data all the way to the root. this is done taking locks. the advantage of this is that if there are 2 or more cpus on different nodes starting grace periods, they will eventually meet, and only one will proceed. moreover, if a cpu lags behind, it can stop from there. when the gp_seq_needed is updated all the way to the root, if a grace period was not already requested, the rcu_gp kthread is woken up

https://elixir.bootlin.com/linux/v6.19.11/source/kernel/rcu/tree.c#L3442
https://elixir.bootlin.com/linux/v6.19.11/source/kernel/rcu/tree.c#L999

#

as i said earlier, the rcu gp kthread works as a state machine, so that if there was no gp already, it is in an init state. from there, it takes notice of cpus that have went online/offline in order to update the rcu_node structures they belong to, and after that, it modifies all the rcu_node structures in order to reset their qs needed bitmaps, and also to set the new value of gp_seq, which, is not taken from gp_seq_needed, but instead, samples a new one from the root

https://elixir.bootlin.com/linux/v6.19.11/source/kernel/rcu/tree.c#L2259
https://elixir.bootlin.com/linux/v6.19.11/source/kernel/rcu/tree.c#L1803

#

the cpus notice that a new grace period has started on the sched clock tick, in the rcu sched clock tick function, where it compares, without taking locks, its gp_seq number of its rcu_data to the gp_seq number of its parent rcu_node. if they dont match, a grace period is started or ended, which is when the rcu core (the per-cpu kthread or softirq) is invoked, which handles beginnings and ends of grace periods, is responsible for the callbacks, and reports quiescence.
if a new grace period is started, rcu_node->core_needs_qs is set. if the rcu tick sees that the cpu is in a quiescent, and core_needs_qs is set, it invokes the rcu core to report this upwards

https://elixir.bootlin.com/linux/v7.0.8/source/kernel/rcu/tree.c#L2696
https://elixir.bootlin.com/linux/v7.0.8/source/kernel/rcu/tree.c#L3710
https://elixir.bootlin.com/linux/v7.0.8/source/kernel/rcu/tree.c#L2835
https://elixir.bootlin.com/linux/v7.0.8/source/kernel/rcu/tree.c#L2498
https://elixir.bootlin.com/linux/v7.0.8/source/kernel/rcu/tree.c#L1273

#

theres also special provisions for handling expedited rcu, handling callbacks which have been pending for too long, cpus that are unfortunate enough to only have small windows of quiescence state which the sched clock tick can't account for

#

and finally, it's in the rcu gp cleanup where the gp_seq is advanced on all nodes, gp_seq_needed is advanced, and possibly a new grace period is started

#

in conclusion: run

prime wraith May 18, 2026, 4:22 PM

#

pearl meteor in conclusion: run

#

ill pin it to read later

carmine token May 19, 2026, 6:02 AM

#

tree RCU is the implementation strategy not a separate RCU flavor

#

Also worth noticing that the fanout of each node is very high so the "tree" is just a single node on all non server CPUs

#

Apparently it's 1024 by default

prime wraith May 19, 2026, 7:43 AM

#

Interesting

prime wraith May 19, 2026, 10:41 AM

#

Cleanest uacpi printf integration of all time chad

void uacpi_kernel_log(uacpi_log_level lvl, const uacpi_char *fmt, ...)
{
    va_list vlist;
    struct nested_printf npf;

    va_start(vlist, fmt);
    npf.fmt = fmt;
    npf.vlist = &vlist;

    print(LOG_LEVEL_PREFIX"%cacpi: %pV", uacpi_log_level_to_syslog(lvl), &npf);

    va_end(vlist);
}

haughty notch May 19, 2026, 10:48 AM

#

cacpi?

prime wraith May 19, 2026, 10:49 AM

#

%c

#

in the output it just results in this:

acpi: starting uACPI, version 4.0.0
acpi: RSDP 0x00000000000F52C0 00000014 v00 (BOCHS )
acpi: RSDT 0x000000003FFE230E 00000038 v01 (BOCHS  BXPC    )
acpi: DSDT 0x000000003FFE0040 000020C6 v01 (BOCHS  BXPC    )
acpi: FACP 0x000000003FFE2106 000000F4 v03 (BOCHS  BXPC    )
acpi: APIC 0x000000003FFE21FA 00000078 v03 (BOCHS  BXPC    )
acpi: HPET 0x000000003FFE2272 00000038 v01 (BOCHS  BXPC    )
acpi: MCFG 0x000000003FFE22AA 0000003C v01 (BOCHS  BXPC    )
acpi: WAET 0x000000003FFE22E6 00000028 v01 (BOCHS  BXPC    )

#

but the log level is stored aside in metadata

haughty notch May 19, 2026, 11:41 AM

#

that's cool

thorny kelp May 19, 2026, 1:32 PM

#

prime wraith but the log level is stored aside in metadata

thanks for the idea trl

prime wraith May 19, 2026, 1:35 PM

#

no problem lol, i stole it myself

prime wraith May 20, 2026, 5:08 PM

#

btw @haughty notch this is what clanker said about IOMMU + x2apic:

Standard PCI MSIs and IO-APIC Redirection Table Entries (RTEs) are fundamentally just MMIO writes to the APIC address space (typically starting at 0xFEE00000). In legacy xAPIC mode, the Local APIC actively claims this MMIO region. When a device DMAs a write to that address, the LAPIC intercepts it and processes the interrupt.

When you switch the processor into x2APIC mode, the LAPIC's MMIO interface is completely disabled in favor of the MSR interface. This creates a routing problem: if a PCI device sends a standard MSI, it issues a memory write to 0xFEExxxxx, but the LAPIC is no longer listening on that physical memory window.

Intel's hardware solution relies entirely on the VT-d IOMMU. The Interrupt Remapping engine intercepts those legacy MMIO writes from devices (or the IO-APIC) and translates them into the proper x2APIC interrupt messages on the system interconnect (QPI/UPI). Without the IR engine enabled, the chipset provides no architecturally guaranteed hardware path to route a legacy memory-mapped MSI to an MSR-only x2APIC.

#

idk if its completely true

ornate jasper May 20, 2026, 5:10 PM

#

what were you asking it? because i didnt think that was true

prime wraith May 20, 2026, 5:10 PM

#

ornate jasper what were you asking it? because i didnt think that was true

why does linux only enable x2apic if there's an iommu (barring paravirt)

ornate jasper May 20, 2026, 5:10 PM

#

does it? i was looking at the x2apic paths and didnt see any iommu checks

#

(looked when i was trying to figure out the edge cases around ap startup with mixed x/x2apic cpu states because of kernel turning on x2apic before ap startup)

prime wraith May 20, 2026, 5:11 PM

#

see https://elixir.bootlin.com/linux/v7.0.1/source/arch/x86/kernel/apic/apic.c#L1900

ornate jasper May 20, 2026, 5:12 PM

#

oh thats a new change

#

in 7.0

#

because i was looking in 6.x and that function didnt exist

prime wraith May 20, 2026, 5:14 PM

#

im looking at 6.x and its there also

ornate jasper May 20, 2026, 5:14 PM

#

maybe im just stupid then lmao

prime wraith May 20, 2026, 5:14 PM

#

idk how old it is honestly

prime wraith May 20, 2026, 5:15 PM

#

prime wraith btw <@350748255929499649> this is what clanker said about IOMMU + x2apic: ``` St...

@inland ridge do u know if this is true?

#

tldr x2apic cant intercept msis without an iommu?

ornate jasper May 20, 2026, 5:17 PM

#

oh i see how i missed this

#

i was looking at the enable x2apic path thats used for APs

pearl meteor May 20, 2026, 5:17 PM

#

#

?

#

ah so yes, you indeed need the iommu?

ornate jasper May 20, 2026, 5:18 PM

#

you need an iommu but you can leave it in compatibility mode by default

#

so the iommu just needs to exist not be tinkered with

#

(but you can only target cpus with x2apic id < 256 using the compat mode)

prime wraith May 20, 2026, 5:19 PM

#

So as long as it exists and you didn't touch it, it will identity map them to the right cpu?

ornate jasper May 20, 2026, 5:20 PM

#

for the cpus that are also reachable in xapic mode yes

prime wraith May 20, 2026, 5:20 PM

#

I mean its not like u can encode other cpus in the MSI dst

ornate jasper May 20, 2026, 5:20 PM

#

yeah exactly

#

if you turn on the iommu interrupt remapping stuff it uses a different message format that lets you do more

prime wraith May 20, 2026, 5:21 PM

#

That makes sense ig

ornate jasper May 20, 2026, 5:22 PM

#

(the remappable message format just encodes an index into the address and data and then uses that handle to look up in the remapping table iirc but i havent looked into that too far yet)

pearl meteor May 20, 2026, 5:22 PM

#

the amd iommu does not have anything like that

#

so, i suppose the bios just sets up the individual device table entries interrupt remapping

#

ill check later

#

well, my iommu does not support x2apic KEKW

ornate jasper May 20, 2026, 5:36 PM

#

my amd cpu doesnt even support x2apic afaik

prime wraith May 20, 2026, 5:38 PM

#

pearl meteor well, my iommu does not support x2apic <:KEKW:913708372329644042>

So x2apic msis dont work?

prime wraith May 20, 2026, 5:39 PM

#

ornate jasper my amd cpu doesnt even support x2apic afaik

Cant u check in dmesg?

pearl meteor May 20, 2026, 5:39 PM

#

prime wraith So x2apic msis dont work?

linux refuses to use the x2apic for that reason

#

May 20 16:01:29 archlinux kernel: APIC: Static calls initialized
May 20 16:01:29 archlinux kernel: ACPI: APIC 0x0000000077223000 0000E8 (v05 _ASUS_ Notebook 01072009 AMI  00010013)
May 20 16:01:29 archlinux kernel: ACPI: Reserving APIC table memory at [mem 0x77223000-0x772230e7]
May 20 16:01:29 archlinux kernel: ACPI: LAPIC_NMI (acpi_id[0xff] high edge lint[0x1])
May 20 16:01:29 archlinux kernel: IOAPIC[0]: apic_id 33, version 33, address 0xfec00000, GSI 0-23
May 20 16:01:29 archlinux kernel: IOAPIC[1]: apic_id 34, version 33, address 0xfec01000, GSI 24-55
May 20 16:01:29 archlinux kernel: APIC: Switch to symmetric I/O mode setup
May 20 16:01:29 archlinux kernel: x2apic: IRQ remapping doesn't support X2APIC mode
May 20 16:01:29 archlinux kernel: ..TIMER: vector=0x30 apic1=0 pin1=2 apic2=-1 pin2=-1
May 20 16:01:29 archlinux kernel: ACPI: Using IOAPIC for interrupt routing
May 20 16:01:29 archlinux kernel: AMD-Vi: Extended features (0x246577efa2254afa, 0x0): PPR NX GT [5] IA GA PC GA_vAPIC
May 20 16:01:29 archlinux kernel: AMD-Vi: Virtual APIC enabled

prime wraith May 20, 2026, 5:39 PM

#

So like, clanker was right then

pearl meteor May 20, 2026, 5:40 PM

#

suck to be me tbh

#

i had already written the code for the x2apic nooo

#

now i have to deal with the apic bullshit

prime wraith May 20, 2026, 5:41 PM

#

pearl meteor ``` May 20 16:01:29 archlinux kernel: APIC: Static calls initialized May 20 16:0...

Can you grep for routing mode

pearl meteor May 20, 2026, 5:41 PM

#

May 20 16:01:29 archlinux kernel: ACPI: Using IOAPIC for interrupt routing
May 20 16:01:29 archlinux kernel: PCI: Using ACPI for IRQ routing

#

?

prime wraith May 20, 2026, 5:42 PM

#

Nah, the apic routing, it should be around the same place

#

Like physical or logical

#

Also why the hell did amd even add an x2apic to that cpu if it cant receive MSIs

#

That makes it like literally useless

pearl meteor May 20, 2026, 5:43 PM

#

idk

prime wraith May 20, 2026, 5:43 PM

#

Unless there's still a hidden compat mode

pearl meteor May 20, 2026, 5:43 PM

#

also idk what message you're talking about

prime wraith May 20, 2026, 5:44 PM

#

Can you paste dmesg?

pearl meteor May 20, 2026, 5:44 PM

#

if i grep for physical or logical there's nothing

#

and those are the only messages with apic

prime wraith May 20, 2026, 5:44 PM

#

What if you grep for phys

pearl meteor May 20, 2026, 5:44 PM

#

nope

prime wraith May 20, 2026, 5:44 PM

#

Or flat

pearl meteor May 20, 2026, 5:44 PM

#

neither

prime wraith May 20, 2026, 5:44 PM

#

Bro

#

Oh well

#

Can you test MSIs on it with some hobby os (or your own)?

#

If they actually never work

pearl meteor May 20, 2026, 5:45 PM

#

yeah, i just started writing the code for them

prime wraith May 20, 2026, 5:45 PM

#

Cool

pearl meteor May 20, 2026, 5:45 PM

#

this conversation rn has striked me a little bit

#

who knows if thanks to it i have saved hours of debugging

prime wraith May 20, 2026, 5:45 PM

#

I mean they should work im 90% sure, otherwise they simply wouldn't add x2apic

#

Unless who knows

pearl meteor May 20, 2026, 5:46 PM

#

ioapic interrupts i get with the x2apic

#

for msis ill have to see

prime wraith May 20, 2026, 5:46 PM

#

Those aren't msis so yeah

#

Lets see tho

pearl meteor May 20, 2026, 5:47 PM

#

i could write a pcie aer driver

#

unironically

#

on every boot i get a correctable error

#

and i think that those are signaled with msis

prime wraith May 20, 2026, 5:47 PM

#

Lol

#

Is your Mobo broken?

pearl meteor May 20, 2026, 5:47 PM

#

unless they're logged and linux just reads them

#

idk

#

to be fair bluetooth does not work properly

#

it disconnects randomly, and the xhci driver complains about electromagnetic interference

#

kernel: usb usb3-port3: disabled by hub (EMI?), re-enabling...

prime wraith May 20, 2026, 5:48 PM

#

Lol

pearl meteor May 20, 2026, 5:48 PM

#

and no fucking idea why

#

i have never tampered anything from my laptop

#

well, today i seem to be lucky, theres no error

#

it comes from the nvme btw

#

although i have never had any issues with it

#

kernel: pcieport 0000:00:02.4: AER: Multiple Correctable error message received from 0000:05:00.0 there, from last boot

haughty notch May 20, 2026, 5:52 PM

#

ornate jasper my amd cpu doesnt even support x2apic afaik

Yeah, I have an AMD CPU with IOMMU but not x2APIC afaik

haughty notch May 20, 2026, 5:53 PM

#

pearl meteor `kernel: usb usb3-port3: disabled by hub (EMI?), re-enabling...`

Does Windows work? I've seen that on my AMD "server" as well

pearl meteor May 20, 2026, 5:53 PM

#

neither

#

when that happens the bluetooth icon disappears

#

it's so incredibly annoying

prime wraith May 20, 2026, 5:54 PM

#

lol

pearl meteor May 20, 2026, 5:54 PM

#

in linux the device just disappears from bluetoothctl and i get fucked anyways because i use wireless headsets

prime wraith May 20, 2026, 5:54 PM

#

but yeah sounds like you have a fucked up laptop

#

which model is it

pearl meteor May 20, 2026, 5:54 PM

#

asus tuf gaming a15

#

tbh cant complain

#

i bought it for 650 euros

prime wraith May 20, 2026, 5:55 PM

#

thats decent ig

#

@pearl meteor APIC: Switched APIC routing to: physical flat I have this for example

pearl meteor May 20, 2026, 5:57 PM

#

kernel: APIC: Switch to symmetric I/O mode setup

prime wraith May 20, 2026, 5:57 PM

#

nah

pearl meteor May 20, 2026, 5:57 PM

#

oh wait

#

no

prime wraith May 20, 2026, 5:57 PM

#

i have that one as well

pearl meteor May 20, 2026, 5:57 PM

#

nope, i dont have anything like that

prime wraith May 20, 2026, 5:57 PM

#

😭

ornate jasper May 20, 2026, 5:59 PM

#

i dont think i can get that info from windows or id give you it lol

prime wraith May 20, 2026, 6:01 PM

#

its fine, doesnt really matter

#

https://elixir.bootlin.com/linux/v6.8.12/source/drivers/iommu/amd/init.c#L3233

#

why the hell does that need a fucking state machine KEKW

pearl meteor May 20, 2026, 6:03 PM

#

lol

ornate jasper May 20, 2026, 6:04 PM

#

lmao

prime wraith May 20, 2026, 6:04 PM

#

static int __init check_ivrs_checksum(struct acpi_table_header *table)
{
    int i;
    u8 checksum = 0, *p = (u8 *)table;

    for (i = 0; i < table->length; ++i)
        checksum += p[i];
    if (checksum != 0) {
        /* ACPI table corrupt */
        pr_err(FW_BUG "IVRS invalid checksum\n");
        return -ENODEV;
    }

    return 0;
}

?????

#

bro u have acpica that does that for u

#

yeah that file contains some very interesting stuff

#

crazy code

ornate jasper May 20, 2026, 6:07 PM

#

wild

haughty notch May 20, 2026, 7:08 PM

#

prime wraith bro u have acpica that does that for u

Inb4 acpica was broken

prime wraith May 20, 2026, 9:58 PM

#

Nice to see this still working after a huge amount of work on early hello world stuff

#

basically now i build my own direct map and don't use it for MMIO or other memory

#

so acpi/smbios/ioapic here is using early mmio helpers

#

a ton of cpu init added

#

pretty large pr

prime wraith May 20, 2026, 9:59 PM

#

prime wraith Nice to see this still working after a huge amount of work on early hello world ...

this is the O3+LTO version for good measure

pale cairn May 21, 2026, 3:27 PM

#

prime wraith Nice to see this still working after a huge amount of work on early hello world ...

Thanks for reminding me that my old dell laptop was vPro compatable and showing me that that exists

ornate jasper May 21, 2026, 3:45 PM

#

its really good

prime wraith May 21, 2026, 3:45 PM

#

ornate jasper its really good

did u set it up also?

ornate jasper May 21, 2026, 3:46 PM

#

yeah im all set up i just havent had time to actually do any os testing with it yet

#

life has been super busy lately

prime wraith May 21, 2026, 3:46 PM

#

ah unlucky

pearl meteor May 21, 2026, 4:27 PM

#

May 21 18:21:01 AtiePC kernel: pcieport 0000:00:01.1: AER: Correctable error message received from 0000:01:00.0
May 21 18:21:01 AtiePC kernel: pcieport 0000:00:01.1: AER: Correctable error message received from 0000:01:00.0
May 21 18:21:01 AtiePC kernel: pcieport 0000:00:01.1: AER: Correctable error message received from 0000:01:00.0
May 21 18:21:01 AtiePC kernel: pcieport 0000:00:01.1: AER: Correctable error message received from 0000:01:00.0

#

from the nvidia gpu gnuslashtroll

prime wraith May 21, 2026, 4:29 PM

#

did u have the chance to test out the MSIs with x2apic on it?

pearl meteor May 21, 2026, 4:30 PM

#

im still on it

#

i expect today or tomorrow to have it finished

#

i think i am going to unironically write an aer driver

prime wraith May 23, 2026, 11:46 AM

#

what the hell is up with github actions

haughty notch May 23, 2026, 12:03 PM

#

prime wraith what the hell is up with github actions

GitHub has had like 84% uptime lately

#

thonk or maybe not

#

But there's an incident with actions https://www.githubstatus.com/history

GitHub Status - Incident History

GitHub's Incident and Scheduled Maintenance History

prime wraith May 23, 2026, 12:05 PM

#

how is it possible to fuck up so often

haughty notch May 23, 2026, 12:05 PM

#

They can't keep up with all of the AI slop

vestal zealot May 23, 2026, 12:06 PM

#

Damn, do I actually need to start backing up shit I had on GH now?

#

Under 90% uptime is fucking insane

haughty notch May 23, 2026, 12:06 PM

#

haughty notch GitHub has had like 84% uptime lately

Idk where I've gotten that number from though

#

But like it's bad

#

It's not on their website

haughty notch May 23, 2026, 12:08 PM

#

vestal zealot Under 90% uptime is fucking insane

So like maybe it's not

vestal zealot May 23, 2026, 12:08 PM

#

Well either way people have been getting more and more antsy about keeping stuff there

haughty notch May 23, 2026, 12:09 PM

#

haughty notch Idk where I've gotten that number from though

I mean I've seen web developers on YT complain about it

vestal zealot May 23, 2026, 12:09 PM

#

And I don't want to be one of the tiny unimportant people that can't recover data if/when shit does go sour

haughty notch May 23, 2026, 12:09 PM

#

vestal zealot Well either way people have been getting more and more antsy about keeping stuff...

Can limine go to something like gitlab or something nobody cares about? trl

haughty notch May 23, 2026, 12:12 PM

#

vestal zealot And I don't want to be one of the tiny unimportant people that can't recover dat...

But idk, the nice thing about git is that you almost certainly have a local copy on your PC, so it's not a big deal

vestal zealot May 23, 2026, 12:13 PM

#

Not of all my projects

#

I sometimes delete stuff I have on GH anyway

haughty notch May 23, 2026, 12:14 PM

#

You can also mirror GitHub and gitlab easily

#

And others

sinful river May 23, 2026, 12:14 PM

#

how is github even profitable

haughty notch May 23, 2026, 12:14 PM

#

It's owned by Microsoft, they don't care

wicked juniper May 23, 2026, 12:14 PM

#

sinful river how is github even profitable

maybe fron the enterprise plan?

sinful river May 23, 2026, 12:15 PM

#

no way a few enterprise plans allow for unlimited storage basically for everyone

wicked juniper May 23, 2026, 12:15 PM

#

it has a few limits i think

haughty notch May 23, 2026, 12:15 PM

#

There's copilot

wicked juniper May 23, 2026, 12:15 PM

#

copilot is fucked

#

and def not profitable

sinful river May 23, 2026, 12:15 PM

#

I know they have github pro and stuff

haughty notch May 23, 2026, 12:15 PM

#

Thought AI is a money furnace

sinful river May 23, 2026, 12:15 PM

#

but idk who really pays that

#

(I have github pro)

wicked juniper May 23, 2026, 12:16 PM

#

i have pro from the student thing XD

sinful river May 23, 2026, 12:16 PM

#

yeah so do I

haughty notch May 23, 2026, 12:16 PM

#

Me too

obtuse sparrow May 23, 2026, 5:36 PM

#

haughty notch Thought AI is a money furnace

They use the dollar bills to feed a furnace in a thermoelectric plant that powers solely the data center

prime wraith May 23, 2026, 11:04 PM

#

$ cloc --vcs=git
     203 text files.
     186 unique files.
      18 files ignored.

github.com/AlDanial/cloc v 1.98  T=5.59 s (33.3 files/s, 3651.3 lines/s)
-------------------------------------------------------------------------------
Language                     files          blank        comment           code
-------------------------------------------------------------------------------
C                               49           1954            783           7892
C/C++ Header                   100           1055            643           4177
Python                           4            420             86           1355
CMake                           21            148             22            945
Assembly                         7             86             49            322
C++                              3             82              4            307
YAML                             1             12              3             50
TOML                             1              3              1             13
-------------------------------------------------------------------------------
SUM:                           186           3760           1591          15061
-------------------------------------------------------------------------------

Biggest gdt init you've seen

#

maybe behind the nadia kernel

pearl meteor May 23, 2026, 11:05 PM

#

how are 1000 lines just cmake?

prime wraith May 23, 2026, 11:05 PM

#

kconfig support mostly

#

symtable embedding with LTO etc, many small things like that add up

latent geode May 23, 2026, 11:07 PM

#

mine is like ~250 loc

prime wraith May 23, 2026, 11:07 PM

#

cmake right?

latent geode May 23, 2026, 11:07 PM

#

yes

prime wraith May 23, 2026, 11:07 PM

#

yeah fair

pearl meteor May 23, 2026, 11:07 PM

#

i am still using the limine barebones template tbh lmao

prime wraith May 23, 2026, 11:07 PM

#

lol

latent geode May 23, 2026, 11:08 PM

#

wait that's not right

#

it's not counting lines in all of the files

prime wraith May 23, 2026, 11:08 PM

#

i mean some features i have: linker script preprocessing, kconfig parsing/lazy reconfiguration, any amount of preliminary kernel relinks to stabilize the symbol table, free-after-init references checking, toolchain selection based on kconfig etc

pearl meteor May 23, 2026, 11:09 PM

#

well that's definitely

latent geode May 23, 2026, 11:09 PM

#

cloc says 745
250 was with tokei

pearl meteor May 23, 2026, 11:09 PM

#

a lot of things

prime wraith May 23, 2026, 11:09 PM

#

im also planning on doing the thing where u parse all used CONFIG_* in a translation unit and replace the config.h dependency on those specific key values

#

so you dont have to recompile the entire kernel if one config value changes

latent geode May 23, 2026, 11:09 PM

#

I have modules, kernel dependencies, kallsyms and the kernel of course

prime wraith May 23, 2026, 11:10 PM

#

yeah so similar size cmake

#

how do u ship the symbol table

latent geode May 23, 2026, 11:10 PM

#

similar to how linux does it

prime wraith May 23, 2026, 11:10 PM

#

do u also support debouncing it with LTO?

latent geode May 23, 2026, 11:11 PM

#

what's that?

prime wraith May 23, 2026, 11:11 PM

#

if u link the kernel without a symbol table, then generate the symbol table, then relink with the filled out symbol table all other symbol offsets will change if u have LTO enabled

#

so u have to redo this step multiple times

latent geode May 23, 2026, 11:11 PM

#

yes I do that

prime wraith May 23, 2026, 11:12 PM

#

ah cool

latent geode May 23, 2026, 11:12 PM

#

link 3 times in total iirc

#

very slow with lto

prime wraith May 23, 2026, 11:12 PM

#

same i think

#

actually no

set(ULTRA_NUM_RELINKS 1)
if (CONFIG_LTO)
    set(ULTRA_NUM_RELINKS 2)
endif ()

#

1 + 2 is 3 yes

#

i can count

latent geode May 23, 2026, 11:13 PM

#

:D

prime wraith May 23, 2026, 11:13 PM

#

u should add kconfig support btw

#

its based af

latent geode May 23, 2026, 11:14 PM

#

I currently just pass a few options with -D

#

config is in my todo list :P

prime wraith May 23, 2026, 11:14 PM

#

yeah i used to do the D thing

#

but its so much nicer to just open up menuconfig/guiconfig and select what u need

#

latent geode May 23, 2026, 11:15 PM

#

prime wraith

why are kvm, vmware and hyperv guest support optional?

#

what do they do?

prime wraith May 23, 2026, 11:16 PM

#

u can disable them so the respective paravirt interface is not used and the thing is treated as bare metal

latent geode May 23, 2026, 11:16 PM

#

hmm

prime wraith May 23, 2026, 11:16 PM

#

latent geode what do they do?

vmware and hyperv offer a bunch of paravirt helpers for various things

pearl meteor May 23, 2026, 11:16 PM

#

prime wraith u should add kconfig support btw

is doing that excruciatingly painful?

prime wraith May 23, 2026, 11:16 PM

#

timekeeping, ipis, smp startup etc

latent geode May 23, 2026, 11:16 PM

#

ah

pearl meteor May 23, 2026, 11:16 PM

#

rn i have a config.h and that's it

prime wraith May 23, 2026, 11:16 PM

#

pearl meteor is doing that excruciatingly painful?

what build system?

pearl meteor May 23, 2026, 11:17 PM

#

makefile

prime wraith May 23, 2026, 11:17 PM

#

hmm lol

#

well

#

you will have to parse the resulting .config file and turn it into key=value pairs

#

other than that i guess not

#

u could see how linux does it i guess

#

since they're using makefiles also obviously

latent geode May 23, 2026, 11:19 PM

#

hand-rolled makefiles my beloved ❤️

latent geode May 23, 2026, 11:20 PM

#

prime wraith ``` $ cloc --vcs=git 203 text files. 186 unique files. 18 files ...

what's c++ code btw

prime wraith May 23, 2026, 11:20 PM

#

test harness stuff

#

i test C code using C++ trl

#

and an assert throws an exception in C code

#

or well, in c++ code that also unwinds C code

latent geode May 23, 2026, 11:21 PM

#

it's kinda cool that you can throw and catch exceptions across different languages

#

iirc for example you can throw in c# and catch in c++ or something like that

prime wraith May 23, 2026, 11:22 PM

#

there are some rules regarding that for C iirc

#

it can be finicky

#

but yeah

#

I have stuff like void malloc_phys_range(uint64_t start, uint64_t size); which then makes that range work via phys_to_virt

#

that makes testing kernel stuff super nice

sinful river May 24, 2026, 12:03 AM

#

pearl meteor is doing that excruciatingly painful?

Kconfig is made to work with make

#

It outputs key=value which you can import directly

prime wraith May 24, 2026, 12:04 AM

#

Waait thats true actually

sinful river May 24, 2026, 12:04 AM

#

I used to do that

#

Just include it directly

obtuse sparrow May 24, 2026, 12:47 AM

#

sinful river It outputs key=value which you can import directly

I doubt it would be a pain to parse these on most build systems

ornate jasper May 24, 2026, 1:08 AM

#

obtuse sparrow I doubt it would be a pain to parse these on most build systems

make in particular lets you just inport it though i think was the point

#

because name=value is valid make syntax iirc

obtuse sparrow May 24, 2026, 1:09 AM

#

Yeah

ornate jasper May 24, 2026, 1:09 AM

#

if you arent using make then rip tho lol

obtuse sparrow May 24, 2026, 1:10 AM

#

Also valid bash syntax but make is just overcomplicated shell scripts meme

ornate jasper May 24, 2026, 1:10 AM

#

well yeah

sinful river May 24, 2026, 1:13 AM

#

ornate jasper if you arent using make then rip tho lol

Meson can also import it, it has a kv module

ornate jasper May 24, 2026, 1:14 AM

#

meson was less than one year old the last time i wrote c or c++ lmao

#

but yeah good to know

pearl meteor May 24, 2026, 9:34 AM

#

@prime wraith

#

i am pleased to announce that msis work perfectly fine on my machine with x2apic and no iommu x2apic support

#

so linux is just bollocks lol

prime wraith May 24, 2026, 9:39 AM

#

lol nice

#

I wonder how the hell they achieve that then

pearl meteor May 24, 2026, 9:51 AM

#

is the linux change recent ?

prime wraith May 24, 2026, 9:55 AM

#

Very old I think

ornate jasper May 24, 2026, 1:05 PM

#

might be a quirk almost then

#

some hardware at the time didn't handle it right or smth

prime wraith May 24, 2026, 2:29 PM

#

nice

pearl meteor May 25, 2026, 5:42 PM

#

there's a bit more on the msi story in amd

#

i don't know how it works on modern chipsets (>2006), but on those where the hypertransport specification was relevant, the parent bridges would catch memory accesses to 0xfee0...., and generate an interrupt message over hypertransport

#

in fact, those bridges have a hypertransport capability in the pci configuration space which still exists on modern chipsets, but is completely read only

#

what is remarkable about that capability is that it allows you to enable and disable generation of hypertransport interrupt messages, and on some chipsets it may allowed you to reprogram the msi address range

#

so it definitely isn't just some 0xfee...... bubbling upwards and making its space to the apic, but it's some pcie/infinity fabric thing working together making it work

#

the iommu specification, in the typical amd style of being vague, doesn't explain what are the other implications of not supporting x2apic, apart from not supporting 32 bit ids and iirc changing the interrupt redirection entry

#

so that's why you might have seen me in the server saying before i tested all of this that writing an iommu driver is not necessary for x2apic

carmine token May 25, 2026, 5:58 PM

#

whether something is "necessary" or not is often arguable anyway

#

people sometimes make it sounds as if x86 was a single uniform architecture when in reality each individual machine often needs specific support at the OS level

#

larger commercial OS vendors have dedicated teams just to enable specific machines

ornate jasper May 25, 2026, 6:02 PM

#

yeah if its any more uniform than other arches thats because of decades of convergent evolution

carmine token May 25, 2026, 6:02 PM

#

yeah

ornate jasper May 25, 2026, 6:02 PM

#

not because its actually uniform

carmine token May 25, 2026, 6:02 PM

#

yeah

#

for example, vmware lists each and every ESXi supported server: https://compatibilityguide.broadcom.com/search?program=server&persona=live&column=partnerName&order=asc

Broadcom Compatibility Guide

Broadcom | VMware | Hardware Compatibility Guide

Find compatibility information for Broadcom and VMware products. Search our database for hardware compatibility details.

#

it doesn't even claim to support arbitrary x86 machines

ornate jasper May 25, 2026, 6:03 PM

#

like theres a whole fuckin bit in the PTEs thats reserved on amd and free-for-software on intel, not even core architectural things like that are truly consistent

#

(its bit 8 on non-final entries, which is the global bit on page/block entries)

#

my desktop has nonstandard serial io port addresses

#

there is no uniformity

pearl meteor May 25, 2026, 6:06 PM

#

any reasonable os should uh, have an acpi interpreter and not probe the standard ones, right meme

eternal sparrow May 25, 2026, 6:16 PM

#

ornate jasper like theres a whole fuckin bit in the PTEs thats reserved on amd and free-for-so...

Even syscall/sysret differs between amd and intel

ornate jasper May 25, 2026, 6:17 PM

#

yeah

prime wraith May 28, 2026, 12:06 PM

#

I wonder if I even need "refined" tsc recalibration later in boot, it seems a 50 ms early calibration yields pretty much almost dead on result, like I get 1859.990.. on a 1860 mhz tsc

#

Maybe ill add that later

haughty notch May 28, 2026, 12:18 PM

#

Just recalibrate against an RTC interrupt every second, then you can detect the unstable timer as well trl

prime wraith May 28, 2026, 12:55 PM

#

💀

prime wraith May 29, 2026, 5:47 PM

#

linux larp is in full effect (I have the [0.0000] thing now)

#

cant wait till i get to smp bringup tbh, that should be fun

wide ether May 29, 2026, 6:16 PM

#

prime wraith linux larp is in full effect (I have the [0.0000] thing now)

i have that too

#

what did you do before

prime wraith May 29, 2026, 6:17 PM

#

well literally nothing because i didnt have time

wide ether May 29, 2026, 6:17 PM

#

ah

prime wraith May 29, 2026, 6:17 PM

#

lmao

stable orchid May 29, 2026, 6:19 PM

#

prime wraith linux larp is in full effect (I have the [0.0000] thing now)

is ultra linux larp

#

(compat)

prime wraith May 29, 2026, 6:19 PM

#

depending on what you mean by larp

#

to an extent it is

carmine token May 29, 2026, 6:20 PM

#

zinna output is still a bit nicer imo because it also has proper log levels

prime wraith May 29, 2026, 6:20 PM

#

show

prime wraith May 29, 2026, 6:20 PM

#

carmine token zinna output is still a bit nicer imo because it also has proper log levels

what do they affect?

#

like color or

stable orchid May 29, 2026, 6:21 PM

#

when osdev competition went from uacpi score to logging look

carmine token May 29, 2026, 6:21 PM

#

~~uacpi score is boring show your mlibc compilation time~~

stable orchid May 29, 2026, 6:21 PM

#

real

carmine token May 29, 2026, 6:22 PM

#

prime wraith like color or

yeah, it makes it easy to discern important and non-important msgs

prime wraith May 29, 2026, 6:22 PM

#

yeah I have that planned also

#

right now the console subsystem is mostly a hack

pearl meteor May 29, 2026, 6:27 PM

#

if you're not using panic() on the very important messages or events you're not doing it right

eternal sparrow May 29, 2026, 6:39 PM

#

I color code by subsystem & log level tags postfixed them

haughty notch May 29, 2026, 6:44 PM

#

panic panic level trl

sinful river May 29, 2026, 6:44 PM

#

Imagine color coding 😎

prime wraith May 29, 2026, 6:50 PM

#

pearl meteor if you're not using panic() on the very important messages or events you're not ...

s/pr_info/pr_emerg chad

pearl meteor May 29, 2026, 7:05 PM

#

lol

prime wraith May 30, 2026, 5:01 PM

#

buddy init... OK

[    0.061726] memory: populating memory map [0xFFFF900000000000 - 0xFFFF90000001C000] for phys [0x0000000000000000 - 0x0000000000800000]
[    0.064399] memory: populating memory map [0xFFFF900000000000 - 0xFFFF900001C00000] for phys [0x0000000000000000 - 0x0000000080000000]
[    0.070628] memory: populating memory map [0xFFFF900003800000 - 0xFFFF900008C00000] for phys [0x0000000100000000 - 0x0000000280000000]
[    0.100684] buddy: initial allocator state below
[    0.101700] buddy: order: [4k  ] [8k  ] [16k ] [32k ] [64k ] [128k] [256k] [512k] [1M  ] [2M  ] [4M  ] 
[    0.103653] buddy: free:   3      4      3      1      3      2      0      2      2      3      2013

wide ether May 30, 2026, 5:35 PM

#

ok buddy

prime wraith May 30, 2026, 5:40 PM

#

triple faults

pearl meteor May 30, 2026, 5:40 PM

#

lol

#

yeah having page sizes above 2mb is definitely larp trl

prime wraith May 30, 2026, 5:41 PM

#

i mean its one constant lmao, i just did whatever

pearl meteor May 30, 2026, 5:41 PM

#

yeah

#

i also have the same ngl

pearl meteor May 30, 2026, 7:48 PM

#

https://docs.kernel.org/cpu-freq/index.html

#

this is the ultimate larp

#

to use a (tm) after the name of the project

#

Ultra(TM)

eternal sparrow May 30, 2026, 7:49 PM

#

Ultra(TM)(TM) would go more hard

#

You just have to trademark the trademarking of ultra

sinful river May 30, 2026, 7:55 PM

#

Linux IS a trademark tho

pearl meteor May 30, 2026, 8:01 PM

#

yeah but it's funny because it's the only place where you see that

#

also it is registered by linus so it should be ®️

prime wraith May 31, 2026, 1:16 PM

#

[    0.001533] memory: populating memory map [0xFF80000000000000 - 0xFF8000000000E000] for phys [0x0000000000000000 - 0x0000000000400000]
[    0.002184] memory: populating memory map [0xFF80000000000000 - 0xFF80000001C00000] for phys [0x0000000000000000 - 0x0000000080000000]
[    0.014967] memory: populating memory map [0xFF80000003800000 - 0xFF80000008C00000] for phys [0x0000000100000000 - 0x0000000280000000]
[    0.079866] buddy: initial allocator state below
[    0.079986] buddy: order: [4k  ] [8k  ] [16k ] [32k ] [64k ] [128k] [256k] [512k] [1M  ] [2M  ] [4M  ] 
[    0.080312] buddy: free:   3      5      4      1      1      2      1      2      2      3      2013  
[    0.080871] alloc: now online, builtin caches listed below
[    0.080995] alloc:   Name         Object Size  Order    Max Empty   
[    0.081161] alloc:   builtin_8    8            0        16          
[    0.081296] alloc:   builtin_16   16           0        16          
[    0.081404] alloc:   builtin_32   32           0        16          
[    0.081533] alloc:   builtin_64   64           0        16          
[    0.081664] alloc:   builtin_128  128          0        16          
[    0.081787] alloc:   builtin_256  256          0        16          
[    0.081895] alloc:   builtin_512  512          0        16          
[    0.082031] alloc:   builtin_1k   1024         0        16          
[    0.082169] alloc:   builtin_2k   2048         0        16          
[    0.082308] alloc:   builtin_4k   4096         0        16          
[    0.082450] alloc:   builtin_8k   8192         1        8

I made the thing use huge pages for vmemmap where possible + added clanker-generated phd memset/memcpy

#

80ms to initialize 8G of memory with TCG on a crappy laptop is pretty good imo

#

128mib takes 10ms

cosmic ibex May 31, 2026, 1:18 PM

#

prime wraith 128mib takes 10ms

is that good?

prime wraith May 31, 2026, 1:18 PM

#

i think 10ms for pretty late kernel init is decent

#

(esp with tcg)

#

phd memset gives a 9x speedup on tcg

#

with O0 at least

eternal sparrow May 31, 2026, 1:20 PM

#

prime wraith phd memset gives a 9x speedup on tcg

is this compared to naive or rep movsb memset?

prime wraith May 31, 2026, 1:20 PM

#

i had a byte-by-byte = 0

eternal sparrow May 31, 2026, 1:21 PM

#

rep movsb also heavily sped me up on tcg. i havent tried phd memset

#

mind testing if you got a moment?

prime wraith May 31, 2026, 1:21 PM

#

well mine is not really phd, its just this

void *memset(void *dest, int ch, size_t count)
{
    void *ret = dest;
    u64 pattern;
    size_t qwords, bytes;

    pattern = 0x0101010101010101ULL * (u8)ch;
    qwords = count >> 3;
    bytes = count & 7;

    asm volatile (
        "rep stosq" : "+D"(dest), "+c"(qwords) : "a"(pattern) : "memory"
    );

    if (bytes) {
        asm volatile (
            "rep stosb" : "+D"(dest), "+c"(bytes) : "a"(pattern) : "memory"
        );
    }

    return ret;
}

prime wraith May 31, 2026, 1:22 PM

#

eternal sparrow mind testing if you got a moment?

testing your kernel?

eternal sparrow May 31, 2026, 1:22 PM

#

nah just using this for memset

#

void* memset(void* dst, int c, size_t n) {
    void* tmp = dst;
    asm volatile("rep stosb" : "+D"(dst), "+c"(n) : "a"(c) : "memory");
    return tmp;
}

prime wraith May 31, 2026, 1:22 PM

#

lets see

eternal sparrow May 31, 2026, 1:22 PM

#

i think it can be faster than stosq in some cases? esp on EMRSB or whatever the cpuid bit is

prime wraith May 31, 2026, 1:23 PM

#

eternal sparrow ```c void* memset(void* dst, int c, size_t n) { void* tmp = dst; asm vol...

at least on tcg i get like 3x slowdown

cosmic ibex May 31, 2026, 1:23 PM

#

eternal sparrow ```c void* memset(void* dst, int c, size_t n) { void* tmp = dst; asm vol...

def faster

eternal sparrow May 31, 2026, 1:24 PM

#

prime wraith at least on tcg i get like 3x slowdown

Zamm

cosmic ibex May 31, 2026, 1:24 PM

#

wait...its slower?

prime wraith May 31, 2026, 1:24 PM

#

the qword + byte version gives 200ms, this gives 600ms

eternal sparrow May 31, 2026, 1:24 PM

#

Less code != faster

prime wraith May 31, 2026, 1:24 PM

#

but its probably because tcg can catch this and direct to phd glibc memset

stable orchid May 31, 2026, 1:24 PM

#

this blud didnt saw glibc impl

eternal sparrow May 31, 2026, 1:25 PM

#

my kernel boot is alr so fast i cant even tell id i used that version PSP1G_pspTrollar

prime wraith May 31, 2026, 1:25 PM

#

does it not slow down if u add more memory?

eternal sparrow May 31, 2026, 1:26 PM

#

oh right lmao

#

i do lazy freelist init so its only paging that would slow it down

#

and it uses 1gb pages

prime wraith May 31, 2026, 1:26 PM

#

eternal sparrow Zamm

on kvm it seems both yield almost the same time

#

but tcg likes qword memset a lot more

cosmic ibex May 31, 2026, 1:27 PM

#

stable orchid this blud didnt saw glibc impl

https://github.com/lattera/glibc/blob/master/string/memset.c
wtf is this glibc

GitHub

glibc/string/memset.c at master · lattera/glibc

GNU Libc - Extremely old repo used for research purposes years ago. Please do not rely on this repo. - lattera/glibc

#

wait

#

this is pretty old

eternal sparrow May 31, 2026, 1:27 PM

#

i can notice like ~2ms diffrence between the two when it stalles on paging init and other?

#

but yeah the qword is better on tcg it seems

prime wraith May 31, 2026, 1:28 PM

#

ill probably keep the qword version since it doesnt degrade bare metal but gives a 3x speedup on tcg

eternal sparrow May 31, 2026, 1:28 PM

#

https://github.com/Mintsuki/tinymembench/commits/master/ may be worth adding here and doing some more benchmarks

prime wraith May 31, 2026, 1:29 PM

#

yeah

eternal sparrow May 31, 2026, 1:29 PM

#

me and mint did some tests with rep stosb and on zen3 and zen1 it was faster than some C impls

#

and slower on zen5

#

it was really funky

prime wraith May 31, 2026, 1:29 PM

#

yeah in general this mem* stuff seems very flakey and random between cpu generations

eternal sparrow May 31, 2026, 1:30 PM

#

I wonder how worth it would be to have many implementations for uarches and then hot patch the function at runtime when you detect it

sinful river May 31, 2026, 1:31 PM

#

Do the Linux static trampoline thing too

#

That is a real PhD memcpy

eternal sparrow May 31, 2026, 1:35 PM

#

sinful river Do the Linux static trampoline thing too

a what now?

wicked juniper May 31, 2026, 1:37 PM

#

prime wraith well mine is not really phd, its just this ```c void *memset(void *dest, int ch,...

this is mine

sinful river May 31, 2026, 1:37 PM

#

eternal sparrow a what now?

They do something to switch out functions at runtime to avoid the overhead of doing an indirect function call

eternal sparrow May 31, 2026, 1:38 PM

#

sinful river They do something to switch out functions at runtime to avoid the overhead of do...

Oh I meant like putting the function into its own section and making sure it’s padded enough. Then replacing the memory directly

prime wraith May 31, 2026, 1:38 PM

#

wicked juniper this is mine

Yeah that seems about right, ideally patched at boot time tho

eternal sparrow May 31, 2026, 1:38 PM

#

And not an indirect function call

eternal sparrow May 31, 2026, 1:38 PM

#

wicked juniper this is mine

I don’t think you need the cld there

prime wraith May 31, 2026, 1:39 PM

#

eternal sparrow I don’t think you need the cld there

You dont yes

#

Its guaranteed by the abi

wicked juniper May 31, 2026, 1:39 PM

#

k, it's gone now

sinful river May 31, 2026, 1:39 PM

#

eternal sparrow Oh I meant like putting the function into its own section and making sure it’s p...

That's essentially what they do

wicked juniper May 31, 2026, 1:39 PM

#

same for this one?

eternal sparrow May 31, 2026, 1:40 PM

#

sinful river That's essentially what they do

It probably isn’t too bad to implement tbh

eternal sparrow May 31, 2026, 1:40 PM

#

wicked juniper same for this one?

rep movsb with reverse direction flag is slow as balls iirc

wicked juniper May 31, 2026, 1:40 PM

#

prime wraith Yeah that seems about right, ideally patched at boot time tho

how would i do this btw

wicked juniper May 31, 2026, 1:40 PM

#

eternal sparrow rep movsb with reverse direction flag is slow as balls iirc

well, it's only used when the mem does overlap

eternal sparrow May 31, 2026, 1:41 PM

#

eternal sparrow It probably isn’t too bad to implement tbh

Though I’d definitely want a bit more build system integration for this with linker sections and other bullshit

prime wraith May 31, 2026, 1:41 PM

#

wicked juniper how would i do this btw

See Linux alternatives

sinful river May 31, 2026, 1:41 PM

#

There's a blog post that explains it

#

It's pretty complex

eternal sparrow May 31, 2026, 1:41 PM

#

wahhgone

#

Maybe another day

#

I got better stuff to worry about regardless

pearl meteor Jun 18, 2026, 6:10 PM

#

dead project xd

prime wraith Jun 18, 2026, 6:11 PM

#

Nah I have a shit ton of downstream changes, just been busy with uacpi and IRL work

prime wraith Jun 21, 2026, 1:03 PM

#

https://pastebin.com/U7eHPDyJ
Look at that! Fully automatic real hardware CI (with only 600 loc of clanker python magic trl )

#

https://pastebin.com/zZkLZajK the script itself for anyone curious (or if u have a laptop with intel amt)

pearl meteor Jun 21, 2026, 1:05 PM

#

very cool tbh

thorny kelp Jun 21, 2026, 1:15 PM

#

prime wraith https://pastebin.com/U7eHPDyJ Look at that! Fully automatic real hardware CI (wi...

We're not on a hypervisor
using 'e9' as the early console

#

KEKW

prime wraith Jun 21, 2026, 1:16 PM

#

lol

#

yeah thats fake, i just forced it to actually write to 0x3080

#

which is the SOL COM on that laptop

thorny kelp Jun 21, 2026, 1:16 PM

#

ah

eternal sparrow Jun 21, 2026, 1:21 PM

#

prime wraith which is the SOL COM on that laptop

Is there a good way to probe for this (besides ACPI I'd presume)?

prime wraith Jun 21, 2026, 1:21 PM

#

yeah, the proper way is either DBG2 or just find the SOL (serial over LAN) PCI device and get the BAR address

#

this laptop doesnt have DBG2 so only method 2 would would

#

but i just hardcode the bar address since i dont have anything like pci yet

#

Some laptops are nicer in that they just capture the normal COM1 output

#

but this one is not

eternal sparrow Jun 21, 2026, 1:26 PM

#

prime wraith this laptop doesnt have DBG2 so only method 2 would would

Ahh makes sense,

#

I still need to finish DBG2 parsing but I have no devices that have it lmao

#

I only needed SPCR for QEMU

prime wraith Jun 21, 2026, 1:27 PM

#

eternal sparrow I still need to finish DBG2 parsing but I have no devices that have it lmao

All relatively recent hw should have one

eternal sparrow Jun 21, 2026, 1:28 PM

#

All mine is zen3 or older so maybe im just unlucky?

prime wraith Jun 21, 2026, 1:28 PM

#

Or maybe disabled by bios

eternal sparrow Jun 21, 2026, 1:28 PM

#

When I get a j*b or goto collage j want to get a zen6 machine

prime wraith Jun 21, 2026, 1:30 PM

#

Same trl

#

Or pantherlake

eternal sparrow Jun 21, 2026, 1:32 PM

#

Honestly nova lake may be good

#

Depending on benchmarks

#

That would have FRED and APX

#

Which I wana support my kernel running with APX

sinful river Jun 21, 2026, 1:46 PM

#

the core ultra plus are really good for the price

eternal sparrow Jun 21, 2026, 8:32 PM

#

sinful river the core ultra plus are really good for the price

I will have to consider that then

#

They are using a new socket again so

#

mmLol

#

That's one of the few reasons to use AMD even then

#

I hope they do commit to the new socket though please

#

😭

prime wraith Jun 23, 2026, 7:19 PM

#

SteamDeck over PXE boot with new hyper

sinful river Jun 23, 2026, 7:19 PM

#

is that linux or ultra

prime wraith Jun 23, 2026, 7:20 PM

#

Ultra lol

sinful river Jun 23, 2026, 7:20 PM

#

can't tell, too much larp

#

what is unattended up to <>

prime wraith Jun 23, 2026, 7:20 PM

#

trl

#

The amount of time the kernel can basically go idle until it needs to "tick" the timekeeper

sinful river Jun 23, 2026, 7:22 PM

#

huh

#

for me PM timer is like 3 mins

prime wraith Jun 23, 2026, 7:22 PM

#

for me its usually 2 seconds

#

this one is 32-bits i guess

sinful river Jun 23, 2026, 7:22 PM

#

oh yeah that oops

#

sorry

prime wraith Jun 23, 2026, 7:23 PM

#

havent seen 32-bit pm timer anywhere before

#

but i have no other amd hardware so

#

well other than my main pc which im not going to bother booting this on

#

..and the ps5

prime wraith Jun 23, 2026, 7:25 PM

#

prime wraith havent seen 32-bit pm timer anywhere before

bro made 32 bit pm timer and didnt bother implementing 64 bit hpet

#

and didnt bother adding cpuid for tsc

eternal sparrow Jun 23, 2026, 7:25 PM

#

prime wraith SteamDeck over PXE boot with new hyper

This makes me want to redo my logs somewhat

prime wraith Jun 23, 2026, 7:27 PM

#

What do u want to change?

thorny kelp Jun 23, 2026, 7:39 PM

#

copy linux entirely

prime wraith Jun 23, 2026, 7:41 PM

#

linux log is a fucking mess if u enable anything > pr_info

#

so i wouldnt go there

#

also it has no color support for any of the backends

eternal sparrow Jun 23, 2026, 7:48 PM

#

prime wraith What do u want to change?

Eh the format and the macros are a bit fucky

#

And not super consistent though shit

#

Probably will be easier to do once everything is finished and I do my log ring shit tbf

prime wraith Jun 23, 2026, 7:49 PM

#

ah

#

yeah the log ring has been super useful even at this stage

#

because this framebuffer console is registered way later after all the allocators are up so i can do vmalloc + wc remap

eternal sparrow Jun 23, 2026, 7:49 PM

#

I just rebuild the limine HHDM

#

So I do always have it at boot

#

But some other logs would be nice to delay

prime wraith Jun 23, 2026, 7:50 PM

#

yeah fair, hyper doesnt map any mmio, even the framebuffer so i dont have that

eternal sparrow Jun 23, 2026, 7:50 PM

#

And I can probably cut some init stages

prime wraith Jun 23, 2026, 7:50 PM

#

for later real hardware CI i will use usb ACM

#

which obviously can only be registered way way later

eternal sparrow Jun 23, 2026, 7:51 PM

#

Because I have one that's for very early initialization that I had to add due to logging shit not being right at boot on some archs

eternal sparrow Jun 23, 2026, 7:51 PM

#

prime wraith which obviously can only be registered way way later

true

prime wraith Jun 23, 2026, 7:51 PM

#

yeah fair

eternal sparrow Jun 23, 2026, 7:51 PM

#

I was planning to just get an old machine

#

With a serial port

#

Which would be good enough as a basic test running

prime wraith Jun 23, 2026, 7:51 PM

#

i want to have a lot of different machines set up for auto CI so just old stuff wont cut it for me

#

and none of my shit has serial ports 😭

eternal sparrow Jun 23, 2026, 7:52 PM

#

True true, I will probably have USB stuff but I think il only support kernel GDB over serial

#

Since it simplifies it so much

#

As every byte can trigger an NMI

#

And be able to hook into everything nicely

prime wraith Jun 23, 2026, 7:53 PM

#

yeah thats nice ig

#

i hope just logs will be enough for me

#

like real pci hw u can just passthrough, and cpu behavior u can basically validate via kvm

#

so hopefully gdb on real hw wont be needed

eternal sparrow Jun 23, 2026, 7:55 PM

#

Fwiw the serial debugging wouldn't be for CI even

#

It would be for needing to debug really stupid bullshit things

prime wraith Jun 23, 2026, 7:55 PM

#

yeah im just saying its probably only really niche things that need debugging that way

#

like if its a driver u can just do a passthrough and do it in on linux

eternal sparrow Jun 23, 2026, 7:56 PM

#

I think it was useful for me once iirc

prime wraith Jun 23, 2026, 7:56 PM

#

and if it dies due to cpu stuff it will most likely die on kvm on that hardware

eternal sparrow Jun 23, 2026, 7:56 PM

#

I forget what I used it for

#

But I do know it did find an issue

prime wraith Jun 23, 2026, 7:56 PM

#

id be interested to know what it was u debugged with it

versed obsidian Jun 23, 2026, 8:07 PM

#

prime wraith SteamDeck over PXE boot with new hyper

it looks beautiful

eternal sparrow Jun 23, 2026, 8:33 PM

#

prime wraith id be interested to know what it was u debugged with it

Iirc it was a deadlock probably

#

Only happened on my machine

#

That or schedular corruption

#

mmLol

prime wraith Jun 24, 2026, 1:18 PM

#

who the hell posted ultra on hacker news

sinful river Jun 24, 2026, 1:26 PM

#

prime wraith who the hell posted ultra on hacker news

What's annoying is that when people do that then you can't post it again

#

Maybe you would've liked to post the repo later? Nope

prime wraith Jun 24, 2026, 1:28 PM

#

wtf

#

i hope its like a comment or something

twilit geode Jun 24, 2026, 1:47 PM

#

sinful river What's annoying is that when people do that then you can't post it again

Huh what

sinful river Jun 24, 2026, 1:47 PM

#

you cant post duplicate links

twilit geode Jun 24, 2026, 1:50 PM

#

Damn

twilit geode Jun 24, 2026, 1:56 PM

#

sinful river you cant post duplicate links

Is that a recent change?

#

I was able to send the astral minecraft link twice last year

sinful river Jun 24, 2026, 2:05 PM

#

twilit geode Is that a recent change?

No always has been

sinful river Jun 24, 2026, 2:06 PM

#

twilit geode I was able to send the astral minecraft link twice last year

Maybe it's cuz the first one didn't get traction

vestal zealot Jun 24, 2026, 2:06 PM

#

sinful river you cant post duplicate links

Like at all? What kind of BS is that?

sinful river Jun 24, 2026, 2:06 PM

#

It makes sense

vestal zealot Jun 24, 2026, 2:06 PM

#

No it doesn't, what if someone just posts slander and now you can't post your own project anymore

sinful river Jun 24, 2026, 2:06 PM

#

There's a threshold of votes you need to get tho

#

but I was pissed off cuz I wanted to post my thing on HN but someone else posted it at the wrong time without asking me first and now I can't repost it cuz the first one got like 20 votes

vestal zealot Jun 24, 2026, 2:08 PM

#

This policy makes no sense

sinful river Jun 24, 2026, 2:08 PM

#

You can email admins apparently but that's too much trouble

sinful river Jun 24, 2026, 2:08 PM

#

vestal zealot This policy makes no sense

It's to avoid reposts

#

Which makes sense

#

But yea it's annoying

vestal zealot Jun 24, 2026, 2:09 PM

#

yeah ok sure but this is too draconian

#

also you can trivially get around it with an URL shortener

sinful river Jun 24, 2026, 2:11 PM

#

That's probably forbidden

#

Osdev hackernews alternative when

#

(that's lobsters)

twilit geode Jun 24, 2026, 2:12 PM

#

sinful river but I was pissed off cuz I wanted to post my thing on HN but someone else posted...

When is the right time so I can repost the astral.wine stuff troll3dflip

sinful river Jun 24, 2026, 2:13 PM

#

Mathew do you have a lobsters account

#

I can hook you up if you want

twilit geode Jun 24, 2026, 2:13 PM

#

I dont

#

Sure

haughty notch Jun 24, 2026, 7:00 PM

#

sinful river Osdev hackernews alternative when

Osdev forums trl

stable orchid Jun 24, 2026, 7:22 PM

#

sinful river Osdev hackernews alternative when

#

https://cdn.discordapp.com/attachments/734354115567812659/1471217554868408416/IMG_0921.gif

thorny kelp Jun 24, 2026, 7:23 PM

#

sinful river Osdev hackernews alternative when

here

prime wraith Jun 24, 2026, 10:23 PM

#

https://news.ycombinator.com/item?id=48650320

mrunix

Ultra: An OS that aims for full ABI compatibility with Linux userland

#

bruh who asked for this

versed obsidian Jun 24, 2026, 10:25 PM

#

@covert raptor did 😭

pearl meteor Jun 24, 2026, 10:38 PM

#

just why

#

new fear unlocked

sinful river Jun 24, 2026, 10:42 PM

#

it would be nice if people did not do that

pearl meteor Jun 24, 2026, 10:43 PM

#

this is a dick move imo

sinful river Jun 24, 2026, 10:43 PM

#

yeah

pearl meteor Jun 24, 2026, 10:43 PM

#

specially considering the person has 100 messages in this server

#

and the project is in a very very early stage

sinful river Jun 24, 2026, 10:43 PM

#

but u might be lucky and be able to repost it since it had few votes

pearl meteor Jun 24, 2026, 10:43 PM

#

(despite being a 20k loc gdt init)

sinful river Jun 24, 2026, 10:43 PM

#

pearl meteor specially considering the person has 100 messages in this server

probably karma farming

#

or something like that

covert raptor Jun 24, 2026, 11:06 PM

#

sinful river it would be nice if people did not do that

Uhh sorry, what's the problem?

covert raptor Jun 24, 2026, 11:07 PM

#

pearl meteor specially considering the person has 100 messages in this server

Ohh... I apologize for this

covert raptor Jun 25, 2026, 8:24 PM

#

@prime wraith I have contacted HN mods and they have deleted the submission now

#

I apologize again for posting it without your consent

prime wraith Jun 25, 2026, 8:30 PM

#

thanks

prime wraith Jun 27, 2026, 12:23 AM

#

edk2$ git show cf9ff46b088a1df9c8ab81f31ef9606826ca7977
commit cf9ff46b088a1df9c8ab81f31ef9606826ca7977
Author: Fu Siyuan <[email protected]>
Date:   Tue Jan 9 15:08:57 2018 +0800

    NetworkPkg: Fix incorrect parameter check in PXE.Mtftp() function.
    
    According to UEFI spec, the PXE.Mtftp() should return invalid parameter if the
    BufferPtr parameter was NULL and the DontUseBuffer parameter was FALSE.
    The DontUseBuffer is only used when perform MTFTP/TFTP read operation.
    
    Contributed-under: TianoCore Contribution Agreement 1.0
    Signed-off-by: Fu Siyuan <[email protected]>
    Reviewed-by: Jiaxin Wu <[email protected]>

diff --git a/NetworkPkg/UefiPxeBcDxe/PxeBcImpl.c b/NetworkPkg/UefiPxeBcDxe/PxeBcImpl.c
index 93f3bfa5ba..9068e0686c 100644
--- a/NetworkPkg/UefiPxeBcDxe/PxeBcImpl.c
+++ b/NetworkPkg/UefiPxeBcDxe/PxeBcImpl.c
@@ -855,11 +855,19 @@ EfiPxeBcMtftp (
       (Filename == NULL) ||
       (BufferSize == NULL) ||
       (ServerIp == NULL) ||
-      ((BufferPtr == NULL) && DontUseBuffer) ||
       ((BlockSize != NULL) && (*BlockSize < PXE_MTFTP_DEFAULT_BLOCK_SIZE))) {
     return EFI_INVALID_PARAMETER;
   }
 
+  if (Operation == EFI_PXE_BASE_CODE_TFTP_READ_FILE ||^M
+      Operation == EFI_PXE_BASE_CODE_TFTP_READ_DIRECTORY ||^M
+      Operation == EFI_PXE_BASE_CODE_MTFTP_READ_FILE ||^M
+      Operation == EFI_PXE_BASE_CODE_MTFTP_READ_DIRECTORY) {^M
+    if (BufferPtr == NULL && !DontUseBuffer) {^M
+      return EFI_INVALID_PARAMETER;^M
+    }^M
+  }^M
+^M

So this is why GET_FILE_SIZE wasnt working on real hardware UEFI

#

Another RetardK2 quirk

#

at least the fix is easy i guess

#

if buffer is null and DONT USE BUFFER, return INVALID_PARAMETER

#

truly genius programming

twilit geode Jun 27, 2026, 12:55 AM

#

asking firmware devs to produce good software

carmine token Jun 27, 2026, 5:25 AM

#

Programming against the uefi spec is useless

#

there are many areas where the spec is incomplete and/or doesn't match actual implementations

#

It's better to just look at why edk2 does in practice

dusky kraken Jun 27, 2026, 7:01 AM

#

prime wraith ``` edk2$ git show cf9ff46b088a1df9c8ab81f31ef9606826ca7977 commit cf9ff46b088a1...

i love the windows file endings in the diff

latent geode Jun 27, 2026, 7:16 AM

#

lol

prime wraith Jun 27, 2026, 9:38 AM

#

dusky kraken i love the windows file endings in the diff

Lmfao

errant temple Jun 27, 2026, 9:49 AM

#

I think it's edk2 policy to use windoze line breaks

dusky kraken Jun 27, 2026, 9:56 AM

#

the old lines in the diff don't have them though

pearl meteor Jun 27, 2026, 11:07 AM

#

wait

#

so, it was implemented, but locked ?

#

lol

prime wraith Jun 27, 2026, 11:12 AM

#

pearl meteor so, it was implemented, but locked ?

well, not locked, just the if check was wrong

#

but yeah it was implemented all along

#

to work around it u have to pass any non-null ptr as buffer

#

ill just pass 0xDEADDECAF

pearl meteor Jun 27, 2026, 11:20 AM

#

lol

prime wraith Jun 27, 2026, 12:12 PM

#

Finally all 3 vPRO laptops I have hooked up to auto-ci

This was a crazy amount of work tbh, from dealing with firmware bugs to dealing with dogshit BIOS UIs
I had to switch to PXE boot because IDE-R is unreliable af on some laptops, that meant I had to add PXE support to hyper UEFI
That wasnt working as it turns out due to an old EDK2 bug where GET_FILE_SIZE requires a non-null useless buffer ptr
Then I realized UEFI wasnt enough because an older vpro laptop I have doesnt support PXE over UEFI so I hadd to add a hyper BIOS PXE implementation as well
Then SOL wasnt working on one laptop so I had to debug and turns out I have to do config space initialization for the sol redirection device myself
Then some laptops work with amtterm, and some need a meshcmd amtterm, basically legacy vs new
Its hell but it works now!

#

I've also ordered an rpi zero w to hook up non vpro laptops i have to EHCI debug port and use that on those

errant temple Jun 27, 2026, 12:17 PM

#

oh wait how does vPRO work

prime wraith Jun 27, 2026, 12:17 PM

#

It offers Intel AMT stuff, which is basically remote management, including power management, serial-over-lan, KVM etc etc

ornate jasper Jun 27, 2026, 12:17 PM

#

errant temple oh wait how does vPRO work

firmware lets you control the computer remotely (password protected)

#

(with an api)

prime wraith Jun 27, 2026, 12:18 PM

#

the api is completely fucked and uncodumented btw

ornate jasper Jun 27, 2026, 12:18 PM

#

yup

prime wraith Jun 27, 2026, 12:18 PM

#

and its quirky af because its all firmware

ornate jasper Jun 27, 2026, 12:18 PM

#

the big feature not listed in the list infy gave is storage redirection. aka you can take your computer's normal storage and have access to it read a disk image over the network instead

#

and its only intel computers and only some

#

its cool af tho

prime wraith Jun 27, 2026, 12:19 PM

#

yeah its called IDE-R, basically one AHCI port is fake and acts as networked storage

#

but it only works on some laptops

#

like 1/3 of the ones i have

ornate jasper Jun 27, 2026, 12:20 PM

#

prime wraith yeah its called IDE-R, basically one AHCI port is fake and acts as networked sto...

i think it also works if your computer has nvme but idk if thats still using IDE-R and just disconnecting the nvme or some other thing

prime wraith Jun 27, 2026, 12:20 PM

#

they have USB-R now, but i wasnt able to get it to work

#

at least boot via it

#

because otherwise i dont need it

#

USB-R should use the same API as IDE-R i think

ornate jasper Jun 27, 2026, 12:21 PM

#

im more interested in how it looks to the guest if your normal storage is nvme

prime wraith Jun 27, 2026, 12:21 PM

#

i dont think NVMe-R is a thing

#

because the way IDE-R works is emulated SCSI requests over network

ornate jasper Jun 27, 2026, 12:22 PM

#

which tbh i could check if i wasnt lazy af since the laptop im typing this on has only nvme drives (and the controller keeps randomly dying and needing a full power cycle before anything can detect it again but i cant get actual info on why because windows explodes if you pull its C drive storage out from under it and without a disk drive nothin be logging but besides the point) and has vPRO

prime wraith Jun 27, 2026, 12:22 PM

#

u can try hooking it up to meshcommander and see what it offers

#

but meshcommander is also abandonware now btw

ornate jasper Jun 27, 2026, 12:23 PM

#

it has the storage redirect

prime wraith Jun 27, 2026, 12:23 PM

#

they only offer proper software to corporations now

ornate jasper Jun 27, 2026, 12:23 PM

#

i checked all the features at some point

prime wraith Jun 27, 2026, 12:23 PM

#

if i had to guess its routed as an emulated usb drive over xhci

ornate jasper Jun 27, 2026, 12:23 PM

#

i just dont know how itll look to the guest because i cant be assed to actually run a guest on it yet

prime wraith Jun 27, 2026, 12:28 PM

#

wdym by guest

errant temple Jun 27, 2026, 12:29 PM

#

prime wraith It offers Intel AMT stuff, which is basically remote management, including power...

oooo this might be useful for managarm hwci, my IVB dell laptop supports it apparently

#

got any pointers for setting that up

prime wraith Jun 27, 2026, 12:29 PM

#

go into bios and click ME Setup

#

then enable Intel AMT

ornate jasper Jun 27, 2026, 12:36 PM

#

you may need to actually set a password and if so it'll fail without telling you why if you don't meet the password requirements which I could only find on a forum somewhere

prime wraith Jun 27, 2026, 12:36 PM

#

yep

#

and it has insane password requirements so i just always use Qwerty1@

errant temple Jun 27, 2026, 12:42 PM

#

I don't see anything about AMT in the BIOS

#

even tho the laptop claims vPro and AMT support

prime wraith Jun 27, 2026, 12:43 PM

#

its under the ME settings

#

its a serparate menu thats not in the general bios settings

errant temple Jun 27, 2026, 12:43 PM

#

I don't see ME settings lol

#

even in the boot selection menu

#

(F12 on the dell laptop)

prime wraith Jun 27, 2026, 12:43 PM

#

when u enter general setup it has like Enter BIOS Setup, Enter Boot Menu, and then Enter ME Settings

#

not in like the big bios ui

errant temple Jun 27, 2026, 12:44 PM

#

I understand that part

ornate jasper Jun 27, 2026, 12:44 PM

#

prime wraith its a serparate menu thats not in the general bios settings

it was under bios settings for my dell laptop

#

well sorta

prime wraith Jun 27, 2026, 12:45 PM

#

oh ok, well leo has a dell also

ornate jasper Jun 27, 2026, 12:45 PM

#

in the bios settings there was a nested menu somewhere near the bottom of the list to turn the whole subsystem on

#

actually im leaving home soon and going to shut down the laptop so ill fuckin reboot into bios and take a potato cam shot of it lmao

errant temple Jun 27, 2026, 12:46 PM

#

I'm currently flashing a newer BIOS, sec

#

that laptop is 12 releases behind meme

ornate jasper Jun 27, 2026, 12:47 PM

#

#

here's where the setting was for me

errant temple Jun 27, 2026, 12:47 PM

#

I don't think I see the Managebility thing at all

#

sadmeme

#

note to self: only buy laptops with vPro in the future

ornate jasper Jun 27, 2026, 12:48 PM

#

maybe updating bios will help lol

prime wraith Jun 27, 2026, 12:48 PM

#

i mean if it says it has one it probably has it hidden somewhere

ornate jasper Jun 27, 2026, 12:48 PM

#

and then this menu option to get into it

errant temple Jun 27, 2026, 12:49 PM

#

oh could it be that it doesn't work in docked mode?

ornate jasper Jun 27, 2026, 12:49 PM

#

that'd be weird but maybe ig

#

mine works plugged in but I don't have a laptop dock so idk about docked mode

errant temple Jun 27, 2026, 12:52 PM

#

rn_image_picker_lib_temp_3333269d-8273-4b16-873c-05ad63f60427.jpg

#

sad leo noises

ornate jasper Jun 27, 2026, 12:52 PM

#

unless you turn on the thing in the bios setup it won't be there

#

so id basically just go through the whole bios setup menu lol

errant temple Jun 27, 2026, 12:53 PM

#

rn_image_picker_lib_temp_e5cc534f-e89f-41ca-91cd-1d951be36838.jpg

ornate jasper Jun 27, 2026, 12:53 PM

#

and you'll need to set an admin password in the bios (not system password that's different) to turn it on I think

errant temple Jun 27, 2026, 12:53 PM

#

oh

#

I don't have that, sec

prime wraith Jun 27, 2026, 12:54 PM

#

what do u have under management?

ornate jasper Jun 27, 2026, 12:54 PM

#

idk if not having that hides it tho

#

I just went through the whole tree menu until I found it tbh

errant temple Jun 27, 2026, 12:56 PM

#

still nothing

#

the product page lied then I assume

ornate jasper Jun 27, 2026, 12:57 PM

#

F

prime wraith Jun 27, 2026, 12:57 PM

#

errant temple the product page lied then I assume

do u have a cpu sticker on your laptop

ornate jasper Jun 27, 2026, 12:58 PM

#

my laptop is also a used enterprise one idk if that matters though (like idk if it needs special firmware for the laptop that mine just already had because enterprise. id normally assume no but idk anymore)

prime wraith Jun 27, 2026, 1:00 PM

#

yeah same

#

iirc normal consumer ones never have vpro

#

@errant temple two easy ways to check: vPRO on the cpu sticker or redirection/SOL device in lspci or dmesg | grep serial

errant temple Jun 27, 2026, 1:05 PM

#

ah ok

#

no dice

#

unfortunate

ornate jasper Jun 27, 2026, 1:05 PM

#

F

prime wraith Jun 27, 2026, 1:05 PM

#

F

#

@ornate jasper if u want i can give u my script and the dnsmasq config

#

honestly thats why enterprise laptops are the best, they usually have the most features like that packed in and no gaming gimmick

#

Also interesting thing I noticed: all AMD hardware has 32-bit pmtimer and 32-bit HPET, while all intel hardware has 24-bit pmtimer and 64-bit HPET

red parcel Jun 27, 2026, 1:16 PM

#

prime wraith Also interesting thing I noticed: all AMD hardware has 32-bit pmtimer and 32-bit...

NOTED

#

That is quite interesting ngl

prime wraith Jun 27, 2026, 1:18 PM

#

and this pattern is the same across all hw, like an AMD C50 laptop thats old dogcrap and steamdeck which is relatively recent amd apu

red parcel Jun 27, 2026, 1:19 PM

#

Maybe on desktop it's different?

prime wraith Jun 27, 2026, 1:19 PM

#

maybe yeah, i cant check from windows i think

thorny kelp Jun 27, 2026, 1:21 PM

#

prime wraith Also interesting thing I noticed: all AMD hardware has 32-bit pmtimer and 32-bit...

i thought most modern hardware had a 64 bit hpet

prime wraith Jun 27, 2026, 1:22 PM

#

nah apparently its just amd vs intel

thorny kelp Jun 27, 2026, 1:22 PM

#

(i don't handle hpet rollback currently in my kernel)

thorny kelp Jun 27, 2026, 1:22 PM

#

prime wraith nah apparently its just amd vs intel

interesting

prime wraith Jun 27, 2026, 1:22 PM

#

not that it matters, who uses hpet if u have tsc

red parcel Jun 27, 2026, 1:22 PM

#

Yeah

thorny kelp Jun 27, 2026, 1:22 PM

#

me KEKW

red parcel Jun 27, 2026, 1:22 PM

#

Smh

thorny kelp Jun 27, 2026, 1:22 PM

#

because i haven't implemented tsc yet lmao

prime wraith Jun 27, 2026, 1:23 PM

#

"implemented"

thorny kelp Jun 27, 2026, 1:23 PM

#

maybe i should work on that

prime wraith Jun 27, 2026, 1:23 PM

#

rdtscp

thorny kelp Jun 27, 2026, 1:23 PM

#

prime wraith "implemented"

true

#

lmao

#

implemented as in support it

#

whatever

prime wraith Jun 27, 2026, 1:23 PM

#

-cpu qemu64,-tsc trl

#

that just clears the cpuid bit tho, tsc still works I checked

thorny kelp Jun 27, 2026, 1:25 PM

#

maybe i could fix this too lmfao

#

(don't really need to tho)

ornate jasper Jun 27, 2026, 1:26 PM

#

prime wraith <@121733282676211714> if u want i can give u my script and the dnsmasq config

sure though if it was clanker made I won't use it directly

prime wraith Jun 27, 2026, 1:26 PM

#

thorny kelp maybe i could fix this too lmfao

just add a VT100 bit to your console and emit color based on that

prime wraith Jun 27, 2026, 1:27 PM

#

ornate jasper sure though if it was clanker made I won't use it directly

yeah the amt_ci.py script is entirely clanker written, but u can rewrite it since u dont like python anyway

sterile kayak Jun 27, 2026, 1:31 PM

#

red parcel Maybe on desktop it's different?

my am4/ryzen 5800x system has a 32 bit hpet

thorny kelp Jun 27, 2026, 1:32 PM

#

same- oh wait i have the same exact cpu

prime wraith Jun 27, 2026, 1:36 PM

#

vPRO and serial-over-lan is a life savior for laptops honestly, otherwise your other options are EHCI-debug-port which requires a special cable with a microcontroller which company that made them went out of business or some board (baglebone black, rpi w zero, etc) which has a device mode usb port which can emulate the said microcontroller in software via a linux module. For xHCI its even worse, it also has a debug port capability but that requires a special cable which has vpwr stripped and only one single company makes such cables

#

not to mention xhci/ehci debug port capability requires a bit more code than just a standard x86 uart

#

https://www.cs.usfca.edu/~cruse/cs698s10/ajayxfer.cpp

prime wraith Jun 27, 2026, 1:41 PM

#

prime wraith vPRO and serial-over-lan is a life savior for laptops honestly, otherwise your o...

https://www.datapro.net/products/usb-3-0-super-speed-a-a-debugging-cable.html

USB 3.0 Super-Speed A/A Debugging Cable -- DataPro

USB 3.0 Super-Speed A/A Debugging Cable

#

https://www.cs.usfca.edu/~cruse/cs698s10/
This page is pretty useful btw

twilit geode Jun 27, 2026, 1:45 PM

#

prime wraith not that it matters, who uses hpet if u have tsc

Real ones use the rtl8169 timer but u aint ready for this convo

prime wraith Jun 27, 2026, 1:45 PM

#

lol

pearl meteor Jun 27, 2026, 1:54 PM

#

or the hda timer

prime wraith Jul 3, 2026, 5:10 PM

#

Not doing any ultra dev atm since i've been focused on the loader too much

#

Every since I added PXE support it got me a bit sidetracked

#

theres just too many things i hated about the boot device detection

#

Also reworked the memory mapper to only map usable ram

#

i have a gigantic amount of code just sitting there that i need to go through commit by commit and review etc

#

Also apparently EDK2 for aarch64 dies on startup with -cpu max so i had to do cortex-a72

#

as a sidenote having pxe boot is so convenient its crazy

#

the only downside is i wish laptops supported pxe over wifi

#

Also hyper will have to be excluded from the no clanker list, since i've been using it quite a lot for this. Lately i just started writing out all apis + functions in a .c file with empty bodies and have clanker fill them out and then just use that as a basis

eternal sparrow Jul 3, 2026, 5:20 PM

#

prime wraith Also hyper will have to be excluded from the no clanker list, since i've been us...

meh thats not the worst use, esp if you already have alot of other code related to it imo

prime wraith Jul 3, 2026, 5:20 PM

#

Claude is pretty decent at this, although it usually generates complete dogshit with code duplication so you have to either prompt it a billion times or rewrite everything yourself

prime wraith Jul 3, 2026, 5:20 PM

#

eternal sparrow meh thats not the worst use, esp if you already have alot of other code related ...

yeah, in a mature code base it's really decent

eternal sparrow Jul 3, 2026, 5:21 PM

#

one time i had it call __builtin_memset instead of memset from string.h that i procide

#

🥀

prime wraith Jul 3, 2026, 5:21 PM

#

yeah lol

#

it loves doing ad-hoc lazy bullshit

eternal sparrow Jul 3, 2026, 5:21 PM

#

i feel like thats somehow more effort

#

when iirc string.h was alr in that file

#

with memcpy used

#

😭

prime wraith Jul 3, 2026, 5:22 PM

#

I usually tell it which files to reference and what apis are already available, e.g. i had to explain to it my string_view etc

#

but yeah lol

eternal sparrow Jul 3, 2026, 5:22 PM

#

Since I only use it for stupid shit it kinda just works if I tell it to just do it 💀

#

I try and code as much as I can myself

#

But so times I CBA

prime wraith Jul 3, 2026, 5:22 PM

#

yeah lol

#

i like coding parts that are interesting to me

#

but unfortunately the design and api layout you still have to do yourself basically always or it will produce really bad results

#

which is not that bad ig

eternal sparrow Jul 3, 2026, 5:23 PM

#

That stuff I do like tbh

prime wraith Jul 3, 2026, 5:24 PM

#

yeah sketching out apis is fun ig

eternal sparrow Jul 3, 2026, 5:24 PM

#

It's the really boring shit when I'm half asleep and don't want to use my little brain lower for

prime wraith Jul 3, 2026, 5:24 PM

#

yee

prime wraith Jul 3, 2026, 7:09 PM

#

Wait Downloading https://ftpmirror.gnu.org/gnu/binutils/binutils-2.45.tar.gz... even this fails now

#

Is there another mirror I can use lol

thorny kelp Jul 3, 2026, 7:19 PM

#

prime wraith Is there another mirror I can use lol

the github mirror maybe

#

https://github.com/coreutils/coreutils

prime wraith Jul 3, 2026, 7:20 PM

#

oh yeah maybe

thorny kelp Jul 3, 2026, 7:20 PM

#

thats what i use

#

oh bruh

#

binutils

#

KEKW

prime wraith Jul 3, 2026, 7:20 PM

#

i really wanna drop gcc support im so tired of it not supporting cross compiling by default

#

i only use it to verify my shit is not ub'ed to work only under clang

thorny kelp Jul 3, 2026, 7:21 PM

#

thorny kelp binutils

there's binutils mirrors on github out there

prime wraith Jul 3, 2026, 7:21 PM

#

yeah yeah same idea

thorny kelp Jul 3, 2026, 7:21 PM

#

but i misread it as coreutils lmao

eternal sparrow Jul 3, 2026, 7:30 PM

#

prime wraith Is there another mirror I can use lol

I just hard code the MIT one

#

I also have a mirror on my git server

prime wraith Jul 3, 2026, 7:44 PM

#

eternal sparrow I just hard code the MIT one

Which one is that

thorny kelp Jul 3, 2026, 7:46 PM

#

prime wraith Wait `Downloading https://ftpmirror.gnu.org/gnu/binutils/binutils-2.45.tar.gz......

btw iirc ftpmirror.gnu.org auto selects servers

#

just try again

#

it might've chosen a dead server

thorny kelp Jul 3, 2026, 7:47 PM

#

eternal sparrow I just hard code the MIT one

and this makes sense

eternal sparrow Jul 3, 2026, 7:48 PM

#

prime wraith Which one is that

i forget 😭

#

ah

#

this is the one i used

#

https://mirrors.ocf.berkeley.edu/gnu/gcc/gcc-${version}/gcc-${version}.tar.xz

prime wraith Jul 3, 2026, 7:52 PM

#

Oh thanks

#

I should make a tier system with various gcc mirrors at this point

eternal sparrow Jul 3, 2026, 7:53 PM

#

heh_face

sterile kayak Jul 3, 2026, 10:40 PM

#

prime wraith Wait `Downloading https://ftpmirror.gnu.org/gnu/binutils/binutils-2.45.tar.gz......

I've noticed that ftpmirror.gnu.org doesn't seem to filter for the selected mirror being up

#

And also sometimes ftpmirror itself is down

prime wraith Jul 3, 2026, 10:45 PM

#

Bruh

thorny kelp Jul 3, 2026, 11:07 PM

#

sterile kayak I've noticed that ftpmirror.gnu.org doesn't seem to filter for the selected mirr...

sometimes it gives me okay mirrors and sometimes it gives me ones with dialup speeds

#

like 70kbps and lower

twilit geode Jul 4, 2026, 2:47 AM

#

ftpmirror on its way to give me a random african mirror instead of a brazilian or even south american one

obtuse sparrow Jul 4, 2026, 4:29 AM

#

ftpmirror giving you a mirror on butan

eternal sparrow Jul 4, 2026, 4:30 AM

#

It would be really nice if they fixed ftp mirror

#

And monitored things for being down

#

To not direct to them

#

And to have countrycode.ftpmirror

#

So you can get one that's actually local

thorny kelp Jul 4, 2026, 12:01 PM

#

eternal sparrow It would be really nice if they fixed ftp mirror

savannah.gnu.org trl

prime wraith Jul 5, 2026, 11:49 AM

#

yea... im stuck with the loader shit for a while longer

#

i have a lot more things id like to fix before i continue with the kernel

#

the loader is 4k loc bigger than the kernel KEKW

#

20k vs 16k atm

#

about to change further tho trl

vestal zealot Jul 5, 2026, 12:45 PM

#

prime wraith the loader is 4k loc bigger than the kernel <:KEKW:913708372329644042>

overengineering much? trl

prime wraith Jul 5, 2026, 12:46 PM

#

like i genuinely dont even think so, that thing doesnt even have a gui

#

thats just how much it takes

vestal zealot Jul 5, 2026, 12:47 PM

#

i'm just joshing ya

prime wraith Jul 5, 2026, 12:47 PM

#

lol

eternal sparrow Jul 5, 2026, 1:10 PM

#

prime wraith yea... im stuck with the loader shit for a while longer

I really want to make a pre kernel at this point and support ultra protocol

prime wraith Jul 5, 2026, 1:11 PM

#

What do you support atm?

#

If its limine, the way that protocol is designed is fully compatible with it and you can support both in one binary like pmos

#

You can dynamically detect how you were booted by checking the magic number

eternal sparrow Jul 5, 2026, 1:16 PM

#

Just limine, but I want a pre kernel anyways

#

It will also give me an excuse to learn elf relocations and other things

#

MB2 in the future later

prime wraith Jul 5, 2026, 1:17 PM

#

ah

vestal zealot Jul 5, 2026, 1:24 PM

#

Is the Ultra loader compatible enough I can use it in place of Limine for shits and giggles?

prime wraith Jul 5, 2026, 1:25 PM

#

Well all the structures are different

vestal zealot Jul 5, 2026, 1:25 PM

#

Didn't Ultra also implement Limine protocol though?

prime wraith Jul 5, 2026, 1:25 PM

#

Nope

vestal zealot Jul 5, 2026, 1:25 PM

#

oh

#

which bootloader was it, then? I recall Limine not being the only impl of its own protocol... ?

prime wraith Jul 5, 2026, 1:26 PM

#

No idea

vestal zealot Jul 5, 2026, 1:26 PM

#

Maybe @little aurora knows?

latent geode Jul 5, 2026, 1:30 PM

#

@loud radish might have been dabbling in that

#

iirc

prime wraith Jul 5, 2026, 1:31 PM

#

it's a pretty gigantic protocol so i cant imagine anything would implement it for fun

#

maybe a small subset or smth

loud radish Jul 5, 2026, 1:31 PM

#

i had a semi broken impl yeah

sinful river Jul 5, 2026, 1:31 PM

#

What I would do is have a prekernel loaded by limine

#

or ultra whatever

loud radish Jul 5, 2026, 1:31 PM

#

prime wraith it's a pretty gigantic protocol so i cant imagine anything would implement it fo...

its really not that bad

prime wraith Jul 5, 2026, 1:32 PM

#

terminal feature trl (ik its removed now)

prime wraith Jul 5, 2026, 1:33 PM

#

loud radish its really not that bad

requirements have grown in the last few revisions, e.g. premapping acpi tables which now requires an acpi walker in the bootloader, and iirc there were a few more things like that

#

its not that bad but yes its bad

#

as in a lot of work

sinful river Jul 5, 2026, 1:36 PM

#

prime wraith requirements have grown in the last few revisions, e.g. premapping acpi tables w...

Don't you need some kind of acpi walker in the bootloader anyway

#

If you wanna do NUMA

#

Well you could do without but it'd be annoying

prime wraith Jul 5, 2026, 1:37 PM

#

what sort of numa would you do in a bootloader

sinful river Jul 5, 2026, 1:37 PM

#

Setting things up for the kernel

prime wraith Jul 5, 2026, 1:37 PM

#

what things?

sinful river Jul 5, 2026, 1:37 PM

#

You tell it its numa nodes

prime wraith Jul 5, 2026, 1:37 PM

#

cant the kernel read SRAT itself

sinful river Jul 5, 2026, 1:37 PM

#

Yeah

#

But it's annoying cuz you have a chicken and egg thing so you use your boot allocator

#

idk

prime wraith Jul 5, 2026, 1:38 PM

#

does limine do numa stuff?

#

i havent really thought about that much yet

sinful river Jul 5, 2026, 1:44 PM

#

prime wraith does limine do numa stuff?

No

latent geode Jul 5, 2026, 1:44 PM

#

literally unusable

sinful river Jul 5, 2026, 1:45 PM

#

tf you mean limine won't help my hobby OS run properly on a dual socket server ship

latent geode Jul 5, 2026, 1:45 PM

#

ikr

loud radish Jul 5, 2026, 1:48 PM

#

numa is fake news

ornate jasper Jul 5, 2026, 4:23 PM

#

sinful river But it's annoying cuz you have a chicken and egg thing so you use your boot allo...

not necessarily - ive got it going ok without the loader telling me numa stuff

sinful river Jul 5, 2026, 4:23 PM

#

Yeah it's like the pfndb, it's doable but annoying

#

Maybe I'll do bootloader stuff soon

ornate jasper Jul 5, 2026, 4:24 PM

#

oh nvm i am having the loader tell me numa stuff

#

i already had acpi walking to make sure tables get mapped tho

#

(or rather i use it to put the FACS register area thing in non-temporary vmem range)

#

having a u32 field in the memory map entries for numa node is all you really need to clean up the chicken and egg thing tho

sinful river Jul 5, 2026, 4:26 PM

#

Yeah that and also info for each cpu

ornate jasper Jul 5, 2026, 4:27 PM

#

i dont have any per-cpu info from my loader

#

its easy enough to discover when you are building percpu structs later

#

the SRAT gives you both but my bootloader only cares about the memory bits

#

you basically just use the node for each memmap entry when building up your memblock allocator and then use that to build up everything else

sterile kayak Jul 5, 2026, 4:29 PM

#

Yeah (fully) implementing the limine protocol is a pretty big task

sinful river Jul 5, 2026, 4:44 PM

#

ornate jasper its easy enough to discover when you are building percpu structs later

like node id of the cpu

#

Might as well do it all in the bootloader idk

ornate jasper Jul 5, 2026, 4:44 PM

#

sinful river like node id of the cpu

yeah just iterate that later, doesnt matter for mm init

#

no need to do in bootloader

sinful river Jul 5, 2026, 4:45 PM

#

I don't like it

ornate jasper Jul 5, 2026, 4:45 PM

#

i think memory map numa nodes is the only thing i do in bootloader and nothing else is even remotely annoying enough to chicken-egg to do in bootloader

sinful river Jul 5, 2026, 4:45 PM

#

I'd rather do it all in the loader

#

idk

ornate jasper Jul 5, 2026, 4:45 PM

#

id rather do as little as possible in the loader personally

sinful river Jul 5, 2026, 4:46 PM

#

Nah, the nicer the environment for the kernel the better

ornate jasper Jul 5, 2026, 4:46 PM

#

different philosophies ig

#

im already doing a lot more in the loader than even windows does lmao

#

most of this is due to having literally no mechanism for stack unwinding in uefi (due to what might be llvm bug might be zig bug i dont think anyone understands it enough to know which yet) though

#

id have to at least double the LOC of my bootloader to get it to trace stacks on issues

#

so its so hard to debug

carmine token Jul 5, 2026, 4:49 PM

#

why not just compile with frame pointers? or is that what you're doing?

sinful river Jul 5, 2026, 4:49 PM

#

ornate jasper im already doing a lot more in the loader than even windows does lmao

Windows does a bunch for the kernel

#

I think

#

idk

ornate jasper Jul 5, 2026, 4:50 PM

#

sinful river Windows does a bunch for the kernel

i literally REd winload.efi it doesnt

sinful river Jul 5, 2026, 4:50 PM

#

ah

ornate jasper Jul 5, 2026, 4:50 PM

#

carmine token why not just compile with frame pointers? or is that what you're doing?

that not working is the llvm-or-zig bug that noones been able to trace back yet

sinful river Jul 5, 2026, 4:50 PM

#

What I'm gonna do is have a prekernel loaded by limine that just puts the kernel in a nice state with all APs started

ornate jasper Jul 5, 2026, 4:51 PM

#

i turn on frame pointers for the efi bootloader and nothing changes or happens and i still dont have frame pointers

#

and loading the pdb from the efi filesystem stuff would probably double the lines of code in my loader

#

(plus idk how to even get the current directory of the current image in efi to load the pdb from)

sinful river Jul 5, 2026, 4:52 PM

#

I don't do unwinding

#

well, dwarf stuff at all

#

I just use frame pointers

#

And have my own symbol table

ornate jasper Jul 5, 2026, 4:53 PM

#

as i mentioned, on efi frame pointers dont work at all for me

#

the option just gets ignored

#

so i still cant unwind with them

latent geode Jul 5, 2026, 4:53 PM

#

sinful river I just use frame pointers

same

#

I should do cfi

sinful river Jul 5, 2026, 4:53 PM

#

ornate jasper as i mentioned, on efi frame pointers dont work at all for me

Who cares, your bootloader shouldn't crash

ornate jasper Jul 5, 2026, 4:53 PM

#

ive got frame pointers and cfi for my kernel

ornate jasper Jul 5, 2026, 4:53 PM

#

sinful river Who cares, your bootloader shouldn't crash

no but its hard to debug when im writing it

#

thats why i keep the loader simple

#

and only do the bare minimum

#

because that way it doesnt crash and i dont have to debug it

sinful river Jul 5, 2026, 4:54 PM

#

Don't write bugs 😎

ornate jasper Jul 5, 2026, 4:54 PM

#

lmao

#

id rather do a prekernel than add more to the loader

sinful river Jul 5, 2026, 4:55 PM

#

I'm not sure how the prekernel should load the kernel and where it should sit though, should it sit at the place where the kernel would otherwise exist? And then the prekernel loads the kernel after that?

ornate jasper Jul 5, 2026, 4:55 PM

#

that question is why i never tried to make a prekernel lol

sinful river Jul 5, 2026, 4:57 PM

#

sinful river I'm not sure how the prekernel should load the kernel and where it should sit th...

@carmine token how does managarm do it?

ornate jasper Jul 5, 2026, 4:57 PM

#

winload.efi sets up the memory map, calls SetVirtualAddressMap on efi, loads a few extra acpi tables if theyre missing, and then loads the kernel and calls main
my bootloader sets up the memory map with bonus numa info, calls SetVirtualAddressMap, and then loads the kernel and calls main

#

oh and if the MP protocol for efi is available, winload uses it to invoke the winbvd instruction on all APs after which the APs are all back in standby mode as if nothing happened

#

why it does that i dont fuckin know but it sure does do that

#

i dont even know what the winbvd instruction does lmao

#

oh and winload builds up a gigantic-ass struct it passes to the kernel

carmine token Jul 5, 2026, 5:05 PM

#

sinful river <@358943271151468555> how does managarm do it?

The prekernel allocates memory for the kernel from the same pool that it passes to the kernel

#

not sure if that answers the question

ornate jasper Jul 5, 2026, 5:06 PM

#

i think was more about where in the address space they go

#

thats the question i had about prekernels anyway

sinful river Jul 5, 2026, 5:07 PM

#

carmine token The prekernel allocates memory for the kernel from the same pool that it passes ...

No, I mean where are the executables located?

carmine token Jul 5, 2026, 5:07 PM

#

in virtual memory?

#

or physical

#

the prekernel maps the kernel to its proper place in virtual memory

ornate jasper Jul 5, 2026, 5:08 PM

#

where in virtual memory is the prekernel then/how are conflicts avoided?

sinful river Jul 5, 2026, 5:08 PM

#

Yea

carmine token Jul 5, 2026, 5:10 PM

#

ah

#

the prekernel is either identity mapped (for all boot protocols except limine) or very high (for limine)

obtuse sparrow Jul 5, 2026, 5:11 PM

#

In my prekernel at least, i start on 32-bit protected mode and the bootloader itself loads both the prekernel as the main executable and the kernel as a module

prime wraith Jul 5, 2026, 5:11 PM

#

ornate jasper winload.efi sets up the memory map, calls SetVirtualAddressMap on efi, loads a f...

Whats the point of SetVirtualAddressMap?

carmine token Jul 5, 2026, 5:11 PM

#

carmine token the prekernel is either identity mapped (for all boot protocols except limine) o...

very high = 0xffffffffe0000000

obtuse sparrow Jul 5, 2026, 5:12 PM

#

obtuse sparrow In my prekernel at least, i start on 32-bit protected mode and the bootloader it...

Then i just set up the virtual mappings, enable paging and jump into the virtual memory address of the kernel