- Make the bootloader multiboot compliant
- Rename vector "copy" function to "duplicate"
- Paging system does not necessairilly allocate continues blocks of real memory This can be a problem if the paging system allocates page tables, I am not sure though
threading is a terrible soup. We will have to change the whole multitasking system because we used to switch on task and now will switch on thread this creates the problem of stacks. Each thread will have its own stack duplicated from the caller thread. now thats all fine and dandy but that means we need to create a system to reliably create stacks with a non-heap address but the data does need to be, like, a different page n stuff. Yea i know it's bad but once this is over we will have (nearly) no problem running SMP stuff.
why do the stacks have to be non heap ??? I don't remember and it would make everything a lot harder
lets just create a new stack and copy it from its contents better yet, let's create a heap specifically for this that would not be secure or something but it would be simple and i care more about that currently
we are going to give a timeslice to a task and then that task gives the individual threads CPU's ? that's simple and very inefficient, .... perfect
okay so i was wrong wth
THREADS ARE NEVER COPIED stupid me thread stacks can very well be in heap addresses because they start at an arbitrary entry point in the program meaning the stack is fresh and without ret calls. Whatever it's still complicated asf
the problem is really the first thread but now that i think abt it i really don't think we should've changed anything oml i could have just left the first thread as it was and just spawning new ones with fresh stacks
HHHHHHHHHH