d5f5648b37
We use cargo vendor to generate a .cargo/config file and the vendor directory. Vendoring allows us to lock our dependencies and to modify them easily from the top level Cargo.toml. We vendor all dependencies, including the crates.io ones, which allows for network isolated builds. Signed-off-by: Samuel Ortiz <sameo@linux.intel.com>
2.7 KiB
ELF Image parsing and loading
The boot process is explained from the following two sides.
Loader side
It follows ELF standard which is specified in elf.rs. The entry header and program headers will be inerpreted, and PT_LOAD segments will be loaded into guest memory.
Where kernel is loaded
There are two ways on deciding where the program segments will be loaded.
- One way is to provide an option and allow vmm to specify where to load the image, considering its memory layout.
- The other way is to load image into phdr.p_paddr by default.
Vmm side
Construct zero page
According to the 64-bit boot protocol, the boot parameters (traditionally known as "zero page") should be setup, including setup_header, e820 table and other stuff. However, ELF has no setup_header, nothing returned from ELF loader could be used to fill boot parameters, vmm is totally responsible for the construction.
Configure vcpu
- RIP, the start offset of guest memory where kernel is loaded, which is returned from loader
- 64 bit mode with paging enabled
- GDT must be configured and loaded
bzImage
The boot process is also explained from the following two sides.
Loader side
What will be returned from loader
bzImage includes two parts, the setup and the compressed kernel. The compressed kernel part will be loaded into guest memory, and the following three parts will returned to vmm by loader.
- The start address of loaded kernel
- The offset of memory where kernel is end of loading
- The setup header begin at the offset 0x01f1 of bzImage, this one is an extra compared to the return of ELF loader.
Where kernel is loaded
The same as ELF image loader, there are two ways for deciding where the compressed kernel will be loaded.
- Vmm specify where to load kernel image.
- Load into code32_start (Boot load address) by default.
Additional checking
As what the boot protocol said, the kernel is a bzImage kernel if the protocol >= 2.00 and the 0x01 bit(LOAD_HIGH) is the loadflags field is set. Add this checking to validate the bzImage.
Vmm side
Construct zero page
While vmm build "zero page" with e820 table and other stuff, bzImage loader will return the setup header to fill the boot parameters. Meanwhile, setup_header.init_size is a must to be filled into zero page, which will be used during head_64.S boot process.
Configure vcpu
- RIP, the start address of loaded 64-bit kernel returned from loader + 0x200. Regarding to the 64-bit boot protocol, kernel is started by jumping to the 64-bit kernel entry point, which is the start address of loaded 64-bit kernel plus 0x200.
- 64 bit mode with paging enabled
- GDT must be configured and loaded