Instead of building nested page tables for the offset region, just
offset map the entire thing into kernel memory with one PDP mapping
1GiB large pages. This is more efficient and avoids the "need a
page table to map in a page table" dependency loop.
2MiB large pages were being used for any large page mapping, but the
page manager doesn't correctly handle them everywhere yet. Now only
allow them for offset pointers (eg MMIO space) that will never be
unmapped.
* Non-blocksize-aligned regions could fail to be found. Have the
bootloader load them aligned.
* Consolidating used frame blocks in the bootstrap means these would
have been impossible to free as address space
* mark_permanent wasn't actually removing blocks from the free list
Removed the frame allocation logic from page_manager and replaced it
with using an instance of frame_allocator instead. This had several
major ripple effects:
- memory_initalize() had to change to support this new world
- Where to map used blocks is now passed as a flag, since blocks don't
track their virtual address anymore
- Instead of the complicated "find N contiguous pages that can be
mapped in with one page table", we now just have the bootloader give
us some (currently 64) pages to use both for tables and scratch
space.
- frame_allocator initialization was split into two steps to allow
mapping used blocks before std::move()ing them over
* Heap manager can now manage non-contiguous blocks of memory (currently
all sized at the max block size only)
* Fix a bug where heap manager would try to buddy-merge max-sized blocks
Previously CPU statue was passed on the stack, but the compiler is
allowed to clobber values passed to it on the stack in the SysV x86 ABI.
So now leave the state on the stack but pass a pointer to it into the
ISR functions.
Under KVM we were hitting what look like out-of-order and/or issues
during initialization when writing to the page tables and then
immediately writing to the mapped memory. Adding a memory barrier and
an io_wait() in memory_bootstrap.cpp fixed it.
Processes can now wait on signals/children/time. There is no clock
currently so "time" is just a monotonically increating tick count. Added
a SLEEP syscall to test this waiting/waking.
The syscall/sysret instructions don't swap stacks. This was bad but
passable until syscalls caused the scheduler to run, and scheduling a
task that paused due to interrupt.
Adding a new (hopefully temporary) syscall interrupt `int 0xee` to allow
me to test syscalls without stack issues before I tackle the
syscall/sysret issue.
Also implemented a basic `pause` syscall that causes the calling process
to become unready. Because nothing can wake a process yet, it never
returns.
- Scheduler now has multiple linked_lists of processes at different
priorities
- Process structure improvements
- scheduler::tick() and scheduler::schedule() separation
Added the cpptoml library (and license), and moved to using that for
the initrd manifest. It's now possible to specify the `executable`
flag for files, and the kernel correctly only launches new processes
for the initrd files marked `executable`.
Now any initrd file is treated like a program image and passed to the
loader to load as a process. Very rudimentary elf loading just allocates
pages, copies sections, and sets the ELF's entrypoint as the RIP to
iretq to.
- Create initrd library to support definitions and loading
- Allow tools compiled for the host machine to be built by wscript
- Create makerd tool to build initrd from manifest
- Move screenfont to initrd, so don't load framebuffer initially
The scheduler's create_process now sets up the stack to iretq into a
load_process function, which will load the process image into memory
from within the process' own virtual memory space. Currently this
loading is just copying the old 'taskA' function from kernel space.
More work on process page tables, including only mapping the last 2 pml4
entries (the highest 1TiB of the address space, ie, kernel space) into a
new table.
Includes the work of actually moving the kernel there, which I had
apparently done in name only previously. Oops.
