d60f8ed8d5
The vm_area objects had a number of issues I have been running into when working on srv.init: - It was impossible to map a VMA, fill it, unmap it, and hand it to another process. Unmapping the VMA in this process would cause all the pages to be freed, since it was removed from its last mapping. - If a VMA was marked with vm_flag::zero, it would be zeroed out _every time_ it was mapped into a vm_space. - The vm_area_open class was leaking its page_tree nodes. In order to fix these issues, the different VMA types all work slightly differently now: - Physical pages allocated for a VMA are now freed when the VMA is deleted, not when it is unmapped. - A knock-on effect from the first point is that vm_area_guarded is now based on vm_area_open, instead of vm_area_untracked. An untracked area cannot free its pages, since it does not track them. - The vm_area_open type now deletes its root page_tree node. And page_tree nodes will delete child nodes or free physical pages in their dtors. - vm_flag::zero has been removed; pages will need to be zeroed out further at a higher level. - vm_area also no longer deletes itself only on losing its last handle - it will only self-delete when all handles _and_ mappings are gone.
The jsix operating system
jsix is a custom multi-core x64 operating system that I am building from scratch. It's far from finished, or even being usable - see the Status and Roadmap section, below.
The design goals of the project are:
-
Modernity - I'm not interested in designing for legacy systems, or running on all hardware out there. My target is only 64 bit architecutres, and modern commodity hardware. Currently that means x64 systems with Nehalem or newer CPUs and UEFI firmware. (See this list for the currently required CPU features.) Eventually I'd like to work on an AArch64 port, partly to force myself to factor out the architecture-dependent pieces of the code base.
-
Modularity - I'd like to pull as much of the system out into separate processes as possible, in the microkernel fashion. A sub-goal of this is to explore where the bottlenecks of such a microkernel are now, and whether eschewing legacy hardware will let me design a system that's less bogged down by the traditional microkernel problems.
-
Exploration - I'm really mostly doing this to have fun learning and exploring modern OS development. Initial feature implementations may temporarily throw away modular design to allow for exploration of the related hardware.
A note on the name: This kernel was originally named Popcorn, but I have since
discovered that the Popcorn Linux project is also developing a kernel with that
name, started around the same time as this project. So I've renamed this kernel
jsix (Always styled jsix or j6, never capitalized) as an homage to L4, xv6,
and my wonderful wife.
Status and Roadmap
The following major feature areas are targets for jsix development:
UEFI boot loader
Done. The bootloader loads the kernel and initial userspace programs, and sets up necessary kernel arguments about the memory map and EFI GOP framebuffer. Possible future ideas:
- take over more init-time functions from the kernel
- rewrite it in Zig
Memory
Virtual memory: Sufficient. The kernel manages virtual memory with a number
of kinds of vm_area objects representing mapped areas, which can belong to
one or more vm_space objects which represent a whole virtual memory space.
(Each process has a vm_space, and so does the kernel itself.)
Remaining to do:
- TLB shootdowns
- Page swapping
Physical page allocation: Sufficient. The current physical page allocator implementation suses a group of block representing up-to-1GiB areas of usable memory as defined by the bootloader. Each block has a three-level bitmap denoting free/used pages.
Multitasking
Sufficient. The global scheduler object keeps separate ready/blocked lists per core. Cores periodically attempt to balance load via work stealing.
User-space tasks are able to create threads as well as other processes.
Several kernel-only tasks exist, though I'm trying to reduce that. Eventually only the timekeeping task should be a separate kernel-only thread.
API
In progress. User-space tasks are able to make syscalls to the kernel via fast SYSCALL/SYSRET instructions.
Major tasks still to do:
- The process initialization protocol needs to be re-built entirely.
- Processes' handles to kernel objects need the ability to check capabilities
Hardware Support
- Framebuffer driver: In progress. Currently on machines with a video device accessible by UEFI, jsix starts a user-space framebuffer driver that only prints out kernel logs.
- Serial driver: To do. Machines without a video device should have a user-space log output task like the framebuffer driver, but currently this is done inside the kernel.
- USB driver: To do
- AHCI (SATA) driver: To do
Building
jsix uses the Ninja build tool, and generates the build files for it with
the configure script. The build also relies on a custom sysroot, which can be
downloaded via the Peru tool, or built locally.
Other build dependencies:
- clang: the C/C++ compiler
- nasm: the assembler
- lld: the linker
- mtools: for creating the FAT image
- curl: if using
perubelow to download the sysroot
The configure script has some Python dependencies - these can be installed via
pip, though doing so in a python virtual environment is recommended.
Installing via pip will also install ninja.
A Debian 11 (Bullseye) system can be configured with the necessary build dependencies by running the following commands from the jsix repository root:
sudo apt install clang lld nasm mtools python3-pip python3-venv
python3 -m venv ./venv
source venv/bin/activate
pip install -r requirements.txt
peru sync
Setting up the sysroot
Running peru sync as in the above section will download and unpack the
toolchain into sysroot.
Compiling the sysroot yourself
If you have CMake installed, runing the scripts/build_sysroot.sh script will
download and build a LLVM toolchain configured for building the sysroot, and
then build the sysroot with it.
Built sysroots are actually stored in ~/.local/lib/jsix/sysroots and installed
in the project dir via symbolic link, so having mulitple jsix working trees or
switching sysroot versions is easy.
Building and running jsix
Once the toolchain has been set up, running the ./configure script (see
./configure --help for available options) will set up the build configuration,
and ninja -C build (or wherever you put the build directory) will actually run
the build. If you have qemu-system-x86_64 installed, the qemu.sh script will
to run jsix in QEMU -nographic mode.
I personally run this either from a real debian amd64 bullseye machine or a windows WSL debian bullseye installation. Your mileage may vary with other setups and distros.