This update significantly decouples the generic kernel code
from the BSP code.
Prior to this patch, the BSP had way too much business logic
that should have always been the generic kernel's concern.
While turning IRQs on or off is something that "sounds critical", it is not
unsafe in these sense of compromising memory safety. Rust's unsafe should be
about memory safety only, hence removing it from certain functions.
- Remove the panic version of the GPIO and UART driver. While they were a neat
idea, it proved tedious to drag them along different tutorials where the
virtual memory situation kept on changing. Actually, not much is lost, since
the benefit was only of theoretical nature until now, since everything is
still single-threaded with NullLocks. It is still possible to re-introduce
them later.
- Refactor driver bringup starting with tutorial 14. Instantiating the drivers
only when we are already capable of using the remapped MMIO address makes the
kernel a lot more robust, and the drivers need not care whether their MMIO
addresses are good to use already or not.
- Use console and irq_manager references from the generic kernel code. This
improves decoupling from the BSP, and is needed as a basis for tutorial 14.
The LLVM assembler apparently causes the .equ directive to create symbols
instead of just a local and temporary variable.
Work around this by using const operands with global_asm!.
This patch refactors big chunks of the memory subsystem code.
Most of all, it does away with the the design being based around the raw pointer
type "*const Page". While raw pointers to an actual page seemed like a
compelling idea, in practice it turned out difficult. Rust feels a bit
inconsistent with respect to raw pointers. While it is safe to create them out
of nowhere (only dereferencing is unsafe), it gets weird when multi-threading
comes into picture.
For example, wrapping them into synchronization primitives caused issues because
they don't implement Send. For this reason, we switch to the PageAddress type
which is based on usize, which makes things a lot easier.
Other changes/benefits include:
- Gets rid of unsafe code in the removed PageSlice type.
- Decouple the translation table code and MMIO VA allocation.
- For the translation table tool, make better use of what the ELF format already
provides with respect to memory segmentation and translation. For example, the
tool now queries the ELF file for VA->PA translations and other segment
attributes. This has also the added benefit of reduced BSP code and more
generic code in the tool.
- Packs rbelftools in the Docker image now (used by translation table tool).
- In tutorials 14/15/16, rearrange the PA and VA layout.
