04:05 <Griwes> CompanionCube: well the nice thing about zfs is that once you have the core features, you can kinda just cherry-pick what you want to support and in what order

04:12 <clever> Griwes: ive also been considering adding zfs to little-kernel

04:24 <klange> how did I end up down this path of writing a debugger...

04:24 <klange> I don't even remember

04:25 <clever> lol

04:26 <clever> klange: if i'm not careful, i'm going to wind up being the maintainer of an entire distro!

04:26 <klange> I think I was just paging through command listings and was like "ah, you know, a system call tracer would be nice, how does strace actually do its thing..."

04:27 <klange> And then I read the manuals for ptrace, figured it actually seems pretty straightforward, and since that's also how gdb does things how about a debugger, too, and...

04:30 <klange> I whipped up a little thing to poke at ld.so's internal data to figure out symbols and loaded libs, covers more ground than examining binaries. Tracebacks would be a nice next step if my base pointers are reliable...

04:31 <klange> My strace does a good bit of fancy argument parsing, as well, and I think I fixed a race in waitpid.

04:31 <klange> I can play Doom with the debugger constantly trapping syscalls and it runs fine. Performance under strace kinda depends on whether the terminal is in truetype or bitmap mode...

04:37 <klange> from before fixing a [different] race: https://klange.dev/s/Screenshot%20from%202021-09-24%2012-31-41.png I was trying to use the tools I had to figure out where player health was stored and poke it...

04:37 <clever> klange: i can picture how i would do that without even ptrace

04:38 <clever> just process_vm_readv over the entire addr space, and search for anything with an int of 88, in writable ram

04:38 <clever> then take some damage, and repeat

04:38 <klange> ptrace is where my 'process_vm_read...' is with the PTRACE_PEEKDATA/PTRACE_POKEDATA options...

04:38 <clever> ah

04:39 <clever> but i may need ptrace to modify it, set a watch point to break on any write?

04:39 <clever> and then change it back

04:39 <clever> or a watchpoint on any attempt to read, and fake the read result

04:40 <clever> read based, feels less racey, but more overhead

04:41 <klange> Need to add breakpoints next, and single step. I read the Intel manual page for that stuff at some point in the last few months...

04:42 <klange> Singlestep is just a bit in one of the flag regs? So I set it on return to userspace in the relevant thread and it'll do one instruction and raise an interrupt?

04:42 <clever> klange: oh, have you seen the kvm api?

04:42 <clever> https://david942j.blogspot.com/2018/10/note-learning-kvm-implement-your-own.html

04:42 <bslsk05> ​david942j.blogspot.com: Play With Capture The Flag: [Note] Learning KVM - implement your own kernel

04:42 <clever> this explains how /dev/kvm works, at the low level

04:43 <clever> so you can then implement either end of it (a vm like qemu, or a kernel supporting kvm)

04:44 <klange> please, no, I have enough on my plate, experimenting with virtualization has no business being on my mind...

04:45 <clever> :D

04:46 <klange> Looks like that article just covers making a vm and building a kernel to run _in_ the vm, not the Linux kernel side of the KVM API and/or how that interacts with virtualization instructions

04:47 <clever> but it also includes the ioctl's used to control /dev/kvm

04:47 <clever> how to allocate ram, set cpu regs, create virtual cpus, and pause/resume

04:47 <clever> once you know the api, you can implement either end

04:52 <clever> klange: ive done the same thing with android-auto before, i found source on github for the headunit (radio) end of the protocol, i then filled in the blanks for the android end

05:04 <zid> "once you know how fopen and fclose work, zfs is easy to implement" :p

05:04 <clever> zid: well, if you already have an fs, with a different api, yes :P

05:04 <clever> ive also got that exact problem

05:05 <clever> LK's fs api, isnt directly compatible with fopen/fread

05:05 <clever> but if i implement the glue between the 2, then lua's existing IO library should just work

05:14 <clever> zid: oh, ive even done it with opengl .....

05:14 <clever> i had an opengl example that just drive the hw thru /dev/mem and mmap

05:14 <clever> and i had an opengl client app

05:14 <clever> fill in the blank....

05:15 <clever> it aint pretty, but it works :P

05:31 <klange> yay i can singlestep https://klange.dev/s/Screenshot%20from%202021-09-24%2014-31-11.png

20:00 <vai> https://www.youtube.com/watch?v=18W_LBNg2_8

20:00 <bslsk05> ​'Creative 3D Blaster VLB: $395 DOS Graphics Card from 1995' by LGR (00:35:06)

20:00 <vai> Vesa bus

20:28 <huranimo> I am not even fucking sure how this could work in any other way than what i described. CMOS has 4incoming wires at least cascaded, cause two transistors gates need 4wires, those make out the state 4states, 2*2 cause the output bus is two wires, they can carry 11 00 01 10, obviously 00 and 11 come when write or read strobe is not active , when write strobe is active and data out like wdata ready signal is not there, the bus takes content, and the

20:28 <huranimo> content can not be anything else than what i described cause it is a tri-state buffer on off and stall

20:28 <huranimo> https://www.electronics-tutorials.ws/logic/logic_9.html

20:28 <bslsk05> ​www.electronics-tutorials.ws: Digital Buffer and the Tri-state Buffer Tutorial

20:37 <huranimo> Yes of course there relies a big vulnerability, with many permutes, that is how they got to my computer, that is why john mcafee did those things. It can be defended somewhat, but if an armor or army of computerized alu units over tcp are put against you, commodity hw can not do nothing but to get comporomised or a servers sholder or leg

20:38 <huranimo> when my father traited me along ith my family i knew i was done, magically enough i survived, but of course when the interest is more than usual they will comporomise your life when family allows, three papparazis are more than 1, and do more damage you know

20:39 <huranimo> when you allocate 60cores to compromise one sevrers shoulder of the service , yeah it will carry success if the server is not something special

20:40 <huranimo> when even my family selled info about me when i was 10 and always geve me up, i was all alone against thousounds of attackers

20:41 <huranimo> i am special , but they went paranoid and over the limit, they were desperate and i was only 1