Mark

trying hard. streaming kernel dev (140+ weeks) - about life, art & systems programming
work: @ableton, ex @trailofbits

[Mark @offlinemark]

New kernel post! This time: /proc/*/mem and how it writes to unwritable mem.

Key takeaway: By walking page tables in software, the kernel can access userspace mem without directly dereferencing pointers it gets from userspace. (No WP/SMAP bits involved). https://t.co/U9IOeyHIc5

[Mark @offlinemark]

@zznop_ Amazing, looking forward to it!

[Mark @offlinemark]

it would be cool if binary ninja had a kernel mode

in userspace it's correct that `hlt` would end execution

but contrary to popular belief, `hlt` in the kernel does not turn the computer off or something

it just sleeps the core until the next interrupt. and then continues!

this is why it's so fun to have a hobby os! you can do ring zero experiments and play with the cpu directly =)

[Mark @offlinemark]

big day at work, congrats to the ableton live team =)

[Ableton @Ableton]

Introducing Live 13. Take the (de)tour on the link in comments.

[Mark @offlinemark]

and the end result:

a fully working integration test for preemption

which opts into having the timer interrupt

and existing tests are undisturbed :)

thx. follow for weekly systems programming content from the trenches 🫡

[Mark @offlinemark]

week 142, streaming OS dev (c++)

topic: preemption integration tests

one of my key principles is strong kernel test infrastructure

this week i continued exploring tests for preemptive multitasking

i found a good solution for per-test kernel configs!

here's how it works:

[Mark @offlinemark]

so that's how it works

i use Snitch tags to annotate each test with strings

This can inform the early test code how to config the kernel

which leads to the holy grail:

- kernel init'd exactly once, no reinit
- test can run, do arbitrary kernel stuff, ret back to framework

[Mark @offlinemark]

@_TylerHillery Now is an incredible time for learning new things

[Mark @offlinemark]

@pamoutaf Congrats!!👏

[Mark @offlinemark]

@anthony_codes lfg

[Mark @offlinemark]

my old laptop :')

i learned so much on this thing

post your laptop stickers!!

[Mark @offlinemark]

@ariccio Ok, then i don't actually understand your original comment in this thread. The brace init for Kernel is intentional. The placement new is intentional. The c style cast for the buffers arg is not intentional and just an artifact of the current prototype, not very important

[Mark @offlinemark]

week 141, streaming OS dev (c++)

topic: preemption tests

one of my key principles is strong kernel test infrastructure

this week i explored integration tests for preempt multitasking

few hobby OSes do this. and I can see why! it's hard

here's my problem solving process:

[Mark @offlinemark]

@ariccio start lifetime as is not correct here. i don't want to simply start a lifetime without running ctor, i want to start a lifetime at an address and run ctor

[Mark @offlinemark]

@ariccio Wait Alex are you confusing the placement new for a c style cast? Just checking.

[Mark @offlinemark]

@ariccio No

[Mark @offlinemark]

@ariccio Hehe