Skip to content

Latest commit

 

History

History
193 lines (156 loc) · 7.6 KB

README.md

File metadata and controls

193 lines (156 loc) · 7.6 KB
Raw

CI Status

  • x86-64 platform

Bootloader

  • Can boot from hard disk, flash drive, or CD/DVD using BIOS or EFI
  • Either MBR -> bootloader in partition boot sector reserved area, or GPT boot to EFI system partition (GPT hybrid format, working MBR present)
  • BIOS boot runs in 32 bit protected mode, EFI boot runs in 64 bit long mode
  • Implemented mostly in C++
  • Custom bootloader, supports FAT32 with LFN and ISO9660 with Joliet boot with unified booloader
  • Filesystem accesses abstracted to use own filesystem implementation and BIOS calls for BIOS boot, or use EFI provided filesystem APIs for EFI
  • Builds page tables and enters ELF64 kernel with CPU fully in long mode
  • Implements SMP trampoline
  • Gets physical memory map from BIOS or EFI
  • Uses memory map throughout, can tolerate arbitrarily fragmented memory map
  • Memory map is handed over to kernel so all bootloader allocations are accounted in the map
  • Enumerates VESA BIOS video modes and configures video card for kernel
  • Implements basic exception handlers
  • Shows loading progress bar
  • Processes position independent executable relocations, allowing the kernel to be loaded to an arbitrary address for kernel address space layout randomization.
  • Implements boot menu
  • Abstracts screen output and keyboard input so it can use BIOS or EFI accordingly

Kernel

  • SMP (supporting Multiprocessor Specification and ACPI)
  • LAPIC timer driven preemptive multithreading
  • Priority-based scheduler
  • Tickless scheduler, variable deadline at timeslice exhaustion or timer expiry
  • Processor affinity
  • Oldest timeslice first, optimizes latency over throughput. Timeslice exhaustion causes the repleished timeslice to become young again.
  • Full SSE/AVX/AVX2/AVX-512 support using fxsave/fxrstor or xsave/xsavec/xsaveopt/xrstor where available
  • Recursive paging
  • Per-CPU small block heap
  • Position independent executable memory model with address randomization
  • Demand paged memory
  • Lazy TLB shootdown
  • Memory protection (no mapping of entire physical memory)
  • Interprocessor TLB shootdown
  • High resolution sleep and usleep implementation
  • RTC
  • Atomics
  • GSBASE based CPU-local storage
  • FSBASE and GSBASE per user thread
  • Interrupt Stack Table support with emergency stacks for critical exceptions
  • MCS locks, Spinlocks, Reader/Writer Spinlocks
  • Mutex, Condition Variable, Reader/Writer locks
  • 8259 or LAPIC/IO-APIC support, multiple IO-APICs supported
  • PCI MSI IRQ support including multiple message capability
  • PCI MSI-X IRQ support including routing IRQs to arbitrary CPUs
  • PCIe ECAM (enhanced configuration access method) MMIO configuration space
  • IRQ sharing
  • Fast PCI enumerator with recursive bridge traversal
  • 8042 legacy keyboard and mouse driver
  • XHCI USB 3.1 driver
  • USB Mouse, Keyboard support
  • USB mass storage support (for flash drives and external hard drives)
  • IRQ driven IDE with bus master DMA support, IORDY, 48-bit addressing, ATAPI, bounce buffering
  • IRQ driven AHCI with native command queueing (NCQ), asynchronous I/O, handles up to 32 concurrent commands per port
  • IRQ driven NVME with per-cpu queues and multiple namespace support
  • virtio-block disk driver
  • virtio-gpu driver with dynamic vm window resize handling
  • FAT32 and ISO9660 filesystems with LFN and Joliet support (with full unicode support, including outside the basic multilingual plane)
  • Virtual block storage device abstraction layer
  • Abstract partition probe with MBR and ISO9660 implementations
  • VFS filesystem absraction layer
  • Generalized I/O device page cache with writeback policy and msync flush
  • Dynamically loaded kernel modules with runtime relocation and linkage of kernel exports
  • Non-temporal memory operations
  • SSE/AVX2 optimized memory operations
  • Framebuffer with offscreen shadow buffer and clipping blitter
  • Try/Catch unhandled exception hooking
  • Port 0xE9 based debug output
  • Abstract NIC interface layer
  • RTL8139 NIC driver
  • Ethernet/ARP/UDP/ICMP/DHCP/IPv4 network implemented (so far)
  • Full-featured printf formatter including floating point support
  • High detail exception information written to debug console
  • Several stress test implementations
  • Extensive debugging support for QEMU and Bochs
  • QtCreator project files for easy editing and kernel debugging
  • Instrumented call trace generation with ncurses viewer
  • KASAN address sanitizer
  • UBSAN (undefined behavior sanitizer)
  • All filesystems, partition parsers, and device drivers in separate dynamically loaded modules
  • Kernel unit test module (for unit testing parts of the kernel)
  • PNG parser
  • Fast system memory to video memory clipped blitter with AVX2, SSE4.1, SSE2 optimizations and a platform independent implementation.
  • Support 16-, 24-, and 32-bit per pixel modes. Specialized fast implemetation internals for common modes, arbitrary pixel format is still handled. No palette support.

Build Instructions

These instructions are not strict. You could name your directories whatever and place them wherever (a ramdrive?). These instructions are meant to be simple.

  1. Make a directory where you want it stored and cd into it

     cd ~/code
 mkdir dgos
 cd dgos

  2. Clone the source repo. This will create a directory called dgos

     git clone https://github.com/doug65536/dgos.git

  3. Make directories for out of source build

     mkdir toolchain
 mkdir build

At this point there should be a dgos directory created by git clone, a toolchain directory, and a build directory in the current directory.

Dependencies

You need the following packages: lemon mtools genisoimage qemu-system-x86

Bootstrap (build toolchain if necessary and run configure)

cd build
../dgos/bootstrap --enable-lto --enable-optimize
make disks -j$(nproc)

Toolchain path

If you run into errors, you may need to run . ./set_toolchain_path (or source set_toolchain_path) to enable configure to pick up the toolchain. It will modify the PATH environment variable to include the toolchain.

Debugging

First, cd to the build directory.

Debug output

Debug output is sent to a fifo in the filesystem, which can be read from forever to watch serial output. To do this you need two terminals, one for building, launching qemu, and a second terminal for monitoring debug output.

Terminals that can split the screen into two panes are great for this. In a second terminal, which I'll call the "debug output terminal", run this command in the build directory:

make monitor-debug-output

It will perpetually print serial output to this terminal. You do not need to touch this window again, just read it when needed.

Switch back to the build terminal.

To quickly just run it, run this command:

make run -j$(nproc)

It will run it with TCG emulation. To run it fast, run this command:

make run-smp-kvm -j$(nproc)

run targets don't wait for a debugger, they just run immediately. To have it wait for you to attach a debugger, use debug instead of run.

There are a very large number of targets to run qemu in various configurations. You can construct a variation by piecing together a choice from the following pattern:

make -j$(nproc) {debug|run|test|testdbg}-{up|smp|numa}-{bios|efi}-{pxe|hd}-{iso|mbr|gpt|hyb}-{none|ahci|ide|nvme|usb|virtio}-{vga|gl}-{kvm|icount|tcg|mttcg}

For example, make -j$(nproc) debug-smp-efi-hd-gpt-ahci-vga-kvm will build a VM that waits for the debugger to attach before booting, with SMP (multiple CPUs), EFI firmware/boot, disk boot (not LAN), GPT partition, AHCI disk controller, standard VGA graphics adapter, running with hardware virtualization under KVM.