summary |
shortlog | log |
commit |
commitdiff |
tree
first ⋅ prev ⋅ next
Kit Rhett Aultman [Tue, 3 Sep 2024 03:01:07 +0000 (23:01 -0400)]
trap/console: centralize UART use under console.rs
This commit refactors the UART reading code so that all UART
interactions go through the MmioSerialPort in console.rs. A previous
commit had the UART interrupt handler reading directly from the UART
because some sort of issue prevents calling CONSOLE.lock() outside of
the console crate. However, it's fine to call this from functions
inside the console crate, so this commit implements a console_read()
function there which the trap vector can then call.
Kit Rhett Aultman [Mon, 2 Sep 2024 23:56:39 +0000 (19:56 -0400)]
console/main/trap: enable interrupts, handle UART interrupt
This commit enables interrupts and configures the RISC-V PLIC to deliver
interrupts for the UART (interrupt 10 for qemu). To do this:
* mstatus and mie are configured for m-mode interrupts (mstatus also
likely has s-mode turned on, but we won't use s-mode until we have
to).
* The MmioSerialPort didn't have its init function called; doing so sets
the interrupt enable bits on the UART.
* Configures the PLIC to deliver interrupt 10 on priority 1.
* Adds code to the exception handler to claim, handle, and complete the
interrupt.
A few notes here:
* There's only one interrupt source right now...the UART. So servicing
the interrupt is done right in the handler. There's no need to
determine any other interrupt sources or worry about timing beyond
this, for now.
* Because the MmioSerialPort was really mostly set up for the benefit of
console output, it's not really set up to be used by the interrupt
handler. The code's a bit messy to ensure we read the pending byte
and just echo it back to the terminal. Future commits should address
the awkwardness going on here.
Kit Rhett Aultman [Mon, 2 Sep 2024 23:45:48 +0000 (19:45 -0400)]
console.rs: write entire strings without a loop
The MmioSerialPort already has a write_str() function, so there's no
need to loop the bytes.
Actually, given that MmioSerialPort also has a fmt::Write trait, it's
possible that we don't actually need to implement it for Console at all.
Kit Rhett Aultman [Mon, 2 Sep 2024 23:43:26 +0000 (19:43 -0400)]
main.rs: minor whitespace cleanup
A prior commit didn't have rustfmt run on it, so this commit is a
catchup on the whitespace nits left behind.
Kit Rhett Aultman [Mon, 2 Sep 2024 23:31:18 +0000 (19:31 -0400)]
main.rs: remove commented-out ecall code
Before the commits supporting interrupts from the UART, it's worth
stopping to clean up a little bit. This commented-out code to execute
ecall (which triggers an exception and thus exercises the exception
handler) isn't really needed going forward, so this commit drops it.
Kit Rhett Aultman [Mon, 2 Sep 2024 17:44:02 +0000 (13:44 -0400)]
main.rs: enable floating point
The trap frame already handles the floating point registers, but
floating point support was not enabled via the mstatus register. This
didn't cause an issue on the Ubuntu 22.04 standard version of qemu, but
one built from nightly source will issue an illegal instruction
exception if floating point registers are accessed without the support
being enabled.
This commit fixes the issue by enabling floating point support from
early initialization.
Kit Rhett Aultman [Sat, 24 Aug 2024 15:30:25 +0000 (11:30 -0400)]
trap, main: implement trap handler
This commit implements a trap handler, which handles a number of basic
exception cases and sets the code up for handling interrupts, which will
be really useful for making a serial console properly without relying on
looping.
Changes:
* trap.rs implements a TrapFrame struct and declares one in memory.
This will allow the code to preserve registers and context without
using the stack.
* trap.S implements the low-level parts of the trap handler, copying the
general-purpose and floating-point registers, plus the trap cause and
other context info, into the TrapFrame, and calls the Rust function
m_trap.
* trap.rs implements m_trap, which performs dispatch based on the
exception code and, if the exception is non-fatal, returns the address
of the PC that the trap returns to (typically PC+4).
* main.rs expands the panic handler to log panic information to serial
out so that fatal exceptions are logged. It also adds the trap vector
init code.
This particular method of performing traps is heavily inspired from
Stephen Marz' blog on bare metal Rust for RISC-V:
http://osblog.stephenmarz.com/ch4.html
Kit Rhett Aultman [Sun, 30 Jun 2024 16:41:08 +0000 (12:41 -0400)]
main: clean up logging
A practice I've used for years is to prefix all logging statements with
my name in all-caps. These provisional messages can then be easily
spotted and either kept (sans prefix) or removed. In my haste, though,
I missed one.
Kit Rhett Aultman [Sun, 30 Jun 2024 16:38:01 +0000 (12:38 -0400)]
heap/linker/main: set up a heap allocator
This commit sets up a heap allocator. Again, I cannot sufficiently
stress the usefulness of Meyer Zinn for getting things this far, and
much of what I needed to know about getting linker symbols propagated
into Rust code comes from:
https://meyerzinn.tech/posts/2023/03/08/p1-printing-and-allocating/
Meyer steps through the process of setting up the linked list allocator.
However, there's a buddy list allocator that can be used as a drop-in
replacement, and since this is a more common allocator scheme, and since
doing my own thing means writing my own code and getting more familiar
with Rust, it seemed a divergence worth taking.
Kit Rhett Aultman [Sun, 30 Jun 2024 05:26:12 +0000 (01:26 -0400)]
README.md: created README
Basic statement of goals, current requirements to run, and some notes.
Kit Rhett Aultman [Sun, 30 Jun 2024 05:00:43 +0000 (01:00 -0400)]
main/console: console abstraction and print macros
This commit does a bit of a cleanup and re-org of the UART code so that
code doesn't need to create and carry around a reference to the UART to
print to the console. This is now available through print!() and
println!()
As with the prior commit, this was heavily informed by Meyer Zinn's
excellent blog posts on bare-metal Rust, specifically this one:
https://meyerzinn.tech/posts/2023/03/08/p1-printing-and-allocating/
The idea here is to basically create a singleton Console object and then
only access it behind the print!() and println!() macros. Since the
macros are designed to be used anywhere, they hide the interactions with
a global singleton which performs the debug outputs.
Even though I likely shouldn't have, I also did a rustfmt to clean up
syntax here. Ah, well.
Kit Rhett Aultman [Mon, 24 Jun 2024 04:06:17 +0000 (00:06 -0400)]
Hello world for qemu serial
This commit establishes the overall project structure needed for doing a bare
metal executive on RISC-V. This currently supports the qemu risc64 emulator
because that is a target I can develop across multiple environments as I have
the free time.
This contains:
* Basic project structure
* rustc cross-compilation to riscv64-unknown-none-elf target
* Cargo integration for running qemu
* linking scripts for the qemu riscv64 virt "board" memory layout
* do-nothing panic handler
* _start function that contains the minimal RISC-V initialization: setting up
the global pointer (gp), reserving a stack, and clearing the bss region
* Very simple use of the uart_16550 crate to print a "Hello world" message
This all came together mostly in an afternoon's work and wouldn't have been
possible without these exceptional resources:
"Running Rust code on RISC-V in Qemu"
https://meyerzinn.tech/posts/2023/03/05/running-rust-code-on-risc-v-in-qemu/
"RISC-V from scratch 2: Hardware layouts, linker scripts, and C runtimes"
https://twilco.github.io/riscv-from-scratch/2019/04/27/riscv-from-scratch-2.html