RISC-V Meaty Skeleton with QEMU virt board
WAIT! Have you read Getting Started, Beginner Mistakes, and some of the related OS theory? |
Difficulty level |
---|
Beginner |
Kernel Designs |
---|
Models |
Other Concepts |
This tutorial assumes you have completed RISC-V Bare Bones on the QEMU virt
board, or alternatively, HiFive-1 Bare Bones. If not, you should complete them first for an overview of how to boot your own operating system on RISC-V. This tutorial is deliberately brief on concepts that have already been covered in the bare bones tutorials and their transitive prerequisites.
The bare bones tutorials provide minimal examples that are not structured to enable sustainable mid- to long-term development of the codebase. This tutorial attempts to rectify that by providing a well-structured project that should serve you well through your OSDev journey with the following features:
- Hierarchical project structure with
make
build system for sustainable mid- to long-term development - Includes
debug
target for debugging with GDB (requires cross-debugger targetingriscv64-elf
) - Basic console output through NS16550A UART
- Convenience wrappers for powering off and rebooting the device
- Working
kprintf
supporting base format specifiers (no floating point support; no sub-specifiers; non
specifier) to facilitateprintf
debugging panic
function for kernel panics
RISC-V Bare Bones
- Main article: RISC-V Bare Bones
It is assumed you have completed RISC-V Bare Bones or another comparable bare bones tutorial. Though not a strict requirement, it is useful for confirming that your development environment works and explaining a few basic things.
We won't be reusing any code from those tutorials though, so throw it away (or save it to an archive) and we'll start over again.
Building a Cross-Compiler
- Main article: GCC Cross-Compiler, Why do I need a Cross Compiler?
You must use a GCC Cross-Compiler in this tutorial as in the RISC-V Bare Bones tutorial, with riscv64-elf
as the target.
You must configure your cross-binutils with the --with-sysroot option, otherwise linking will mysteriously fail with the this linker was not configured to use sysroots error message. If you forgot to configure your cross-binutils with that option, you'll have to rebuild it, but you can keep your cross-gcc.
Building a Cross-Debugger (optional)
- Main article: GDB
If you wish to debug your kernel with GDB, you'll need to build it separately with riscv64-elf
as the target, since it's not included by default with GCC and Binutils. Otherwise, if printf
debugging is your style, you may safely skip this section.
The process is similar to building a cross-GCC or cross-binutils, and you may refer to the GDB page in BLFS for most details sans cross-debugging support, but we'll go through the process in detail here anyway.
Fetch the latest version of GDB through https://ftp.gnu.org/gnu/gdb/. The latest version at the time of writing is 12.1:
export GDB_VERSION="12.1"
wget https://ftp.gnu.org/gnu/gdb/gdb-${GDB_VERSION}.tar.xz
Unpack the archive:
tar xvf gdb-${GDB_VERSION}.tar.xz
Now move into the source directory:
pushd gdb-${GDB_VERSION}/
Create a build directory and move into it:
mkdir build
pushd build/
Now export a few variables as with building a cross-GCC or cross-binutils:
export PREFIX="$HOME/opt/cross"
export TARGET=riscv64-elf
export PATH="$PREFIX/bin:$PATH"
Configuration options are mostly the same as with building cross-GCC or cross-binutils. In particular, you may wish to enable the following features:
--enable-tui=yes
: Enables TUI mode for debugging. Requires development headers for Ncurses to be installed on the build host--with-expat
: Build with Expat, a library for XML parsing. Requires development headers for Expat to be installed on the build host
../configure --target=$TARGET \
--prefix="$PREFIX" \
--with-sysroot \
--disable-nls \
--disable-werror \
--enable-languages=c,c++ \
--without-headers \
--enable-tui=yes \
--with-expat
Now build the source code (this can take a while):
make -j$(nproc)
And install:
make -C gdb install
If you wish to keep the source tree available for conveniently re-building GDB in the future (e.g. with a different set of options), clean up the build files now:
make clean
Move one level up to the project root:
popd
Remove the build/
directory we created:
rm -rf build/
Move up one more level:
popd
Double check our cross-debugger is properly installed:
which -- $TARGET-gdb || echo $TARGET-gdb is not in the PATH
Now you may safely remove the source tree and archive, if you wish:
rm -rf gdb-${GDB_VERSION}/
rm gdb-${GDB_VERSION}.tar.xz
If $PREFIX/bin
is not in your path already, you may wish to persist it by writing it to your profile:
echo "export PATH=\"\$HOME/opt/cross/bin:\$PATH\"" >> $HOME/.profile
source $HOME/.profile
Dependencies
You will need these dependencies in order to complete the tutorial:
- QEMU full-system emulator for 64-bit RISC-V, of which your distribution-provided package should suffice
riscv64-elf
toolchain, as discussed above- (Optional, required for debugging with GDB) GNU debugger targeting
riscv64-elf
, as discussed above
This tutorial requires a GNU/Linux system, or a similar enough system. The BSD systems may almost work. OS X is not supported but can possibly be made to work with some changes. Windows is not supported, but Windows environments like Cygwin and Windows Subsystem For Linux (WSL) might work.
Acknowledgements
The initial project setup is based on RISC-V from scratch (C runtime, linker script) and The Adventures of OS (overall project structure, Makefile), with the section on building a cross-GDB based on (Beyond) Linux From Scratch.
Source Code
Fetch the source code from the v0.0.1
release of https://github.com/DonaldKellett/marvelos code-named "Meaty Skeleton", using Git:
git clone --branch v0.0.1 https://github.com/DonaldKellett/marvelos.git
Operating systems development is about being an expert. Take the time to read the code carefully through and understand it. Please seek further information and help if you don't understand aspects of it. This code is minimal and almost everything is done deliberately, often to pre-emptively solve future problems.
Project Structure
Makefile
: used for building the project withmake
build systemmisc/riscv64-virt.dts
: device tree of QEMU RISC-Vvirt
board. See Finding our stack for detailssrc/
: source tree containing files used to build the projectsrc/asm/crt0.s
: minimal C runtime; see Stop! Runtime! for detailssrc/lds/riscv64-virt.ld
: custom linker script adapted from the output ofriscv64-unknown-elf-ld --verbose
; see Link it up for detailssrc/common/
: common utilities and library functions for use across our projectsrc/syscon/
: syscon drivers for poweroff and rebootsrc/uart/
: UART drivers and I/O-related codesrc/kmain.c
: Entry point for our kernel
Makefile
The Makefile
makes extensive use of environment variables for conciseness and configurability. The following top-level targets are supported (along with sub-targets for building only a specific subsystem):
all
: to build the kernel imagerun
: to build the kernel image and run it in QEMUdebug
: to build the kernel image and run it in debug mode to enable debugging with GDB. The "remote" GDB server runs at port$GDB_PORT
on the host (default:1234
). Requires a cross-debugger to be installed; see above for detailsclean
: to clean up build files so the next build won't be affected by the previous build
Run make $TARGET
with TARGET
set to your desired target to execute the specified target. For example, to build the kernel image and run it in QEMU:
make run
Additionally, the default target is all
if not specified:
make
# Build
CC=riscv64-elf-gcc
CFLAGS=-ffreestanding -nostartfiles -nostdlib -nodefaultlibs
CFLAGS+=-g -Wl,--gc-sections -mcmodel=medany
RUNTIME=src/asm/crt0.s
LINKER_SCRIPT=src/lds/riscv64-virt.ld
KERNEL_IMAGE=kmain
# QEMU
QEMU=qemu-system-riscv64
MACH=virt
MEM=128M
RUN=$(QEMU) -nographic -machine $(MACH) -m $(MEM)
RUN+=-bios none -kernel $(KERNEL_IMAGE)
# QEMU (debug)
GDB_PORT=1234
all: uart syscon common kmain
$(CC) *.o $(RUNTIME) $(CFLAGS) -T $(LINKER_SCRIPT) -o $(KERNEL_IMAGE)
uart:
$(CC) -c src/uart/uart.c $(CFLAGS) -o uart.o
syscon:
$(CC) -c src/syscon/syscon.c $(CFLAGS) -o syscon.o
common:
$(CC) -c src/common/common.c $(CFLAGS) -o common.o
kmain:
$(CC) -c src/kmain.c $(CFLAGS) -o kmain.o
run: all
$(RUN)
debug: all
$(RUN) -gdb tcp::$(GDB_PORT) -S
clean:
rm -vf *.o
rm -vf $(KERNEL_IMAGE)
Kernel source
src/common/common.h
#ifndef COMMON_H
#define COMMON_H
int toupper(int);
void panic(const char *, ...);
#endif
src/common/common.c
#include <stdarg.h>
#include "common.h"
#include "../uart/uart.h"
int toupper(int c) {
return 'a' <= c && c <= 'z' ? c + 'A' - 'a' : c;
}
void panic(const char *format, ...) {
kputs("Kernel panic!");
kputs("Reason:");
va_list arg;
va_start(arg, format);
kvprintf(format, arg);
va_end(arg);
asm volatile ("wfi");
}
src/syscon/syscon.h
#ifndef SYSCON_H
#define SYSCON_H
// "test" syscon-compatible device is at memory-mapped address 0x100000
// according to our device tree
#define SYSCON_ADDR 0x100000
void poweroff(void);
void reboot(void);
#endif
src/syscon/syscon.c
#include <stdint.h>
#include "syscon.h"
#include "../uart/uart.h"
void poweroff(void) {
kputs("Poweroff requested");
*(uint32_t *)SYSCON_ADDR = 0x5555;
}
void reboot(void) {
kputs("Reboot requested");
*(uint32_t *)SYSCON_ADDR = 0x7777;
}
src/uart/uart.h
#ifndef UART_H
#define UART_H
#include <stddef.h>
#include <stdarg.h>
// 0x10000000 is memory-mapped address of UART according to device tree
#define UART_ADDR 0x10000000
void uart_init(size_t);
int kputchar(int);
int kputs(const char *);
void kvprintf(const char *, va_list);
void kprintf(const char *, ...);
#endif
src/uart/uart.c
#include <stddef.h>
#include <stdint.h>
#include <stdarg.h>
#include <limits.h>
#include "uart.h"
#include "../common/common.h"
#define to_hex_digit(n) ('0' + (n) + ((n) < 10 ? 0 : 'a' - '0' - 10))
/*
* Initialize NS16550A UART
*/
void uart_init(size_t base_addr) {
volatile uint8_t *ptr = (uint8_t *)base_addr;
// Set word length to 8 (LCR[1:0])
const uint8_t LCR = 0b11;
ptr[3] = LCR;
// Enable FIFO (FCR[0])
ptr[2] = 0b1;
// Enable receiver buffer interrupts (IER[0])
ptr[1] = 0b1;
// For a real UART, we need to compute and set the baud rate
// But since this is an emulated UART, we don't need to do anything
//
// Assuming clock rate of 22.729 MHz, set signaling rate to 2400 baud
// divisor = ceil(CLOCK_HZ / (16 * BAUD_RATE))
// = ceil(22729000 / (16 * 2400))
// = 592
//
// uint16_t divisor = 592;
// uint8_t divisor_least = divisor & 0xFF;
// uint8_t divisor_most = divisor >> 8;
// ptr[3] = LCR | 0x80;
// ptr[0] = divisor_least;
// ptr[1] = divisor_most;
// ptr[3] = LCR;
}
static void uart_put(size_t base_addr, uint8_t c) {
*(uint8_t *)base_addr = c;
}
int kputchar(int character) {
uart_put(UART_ADDR, (uint8_t)character);
return character;
}
static void kprint(const char *str) {
while (*str) {
kputchar((int)*str);
++str;
}
}
int kputs(const char *str) {
kprint(str);
kputchar((int)'\n');
return 0;
}
// Limited version of vprintf() which only supports the following specifiers:
//
// - d/i: Signed decimal integer
// - u: Unsigned decimal integer
// - o: Unsigned octal
// - x: Unsigned hexadecimal integer
// - X: Unsigned hexadecimal integer (uppercase)
// - c: Character
// - s: String of characters
// - p: Pointer address
// - %: Literal '%'
//
// None of the sub-specifiers are supported for the sake of simplicity.
// The `n` specifier is not supported since that is a major source of
// security vulnerabilities. None of the floating-point specifiers are
// supported since floating point operations don't make sense in kernel
// space
//
// Anyway, this subset should suffice for printf debugging
void kvprintf(const char *format, va_list arg) {
while (*format) {
if (*format == '%') {
++format;
if (!*format)
return;
switch (*format) {
case 'd':
case 'i':
{
int n = va_arg(arg, int);
if (n == INT_MIN) {
kprint("-2147483648");
break;
}
if (n < 0) {
kputchar('-');
n = ~n + 1;
}
char lsh = '0' + n % 10;
n /= 10;
char buf[9];
char *p_buf = buf;
while (n) {
*p_buf++ = '0' + n % 10;
n /= 10;
}
while (p_buf != buf)
kputchar(*--p_buf);
kputchar(lsh);
}
break;
case 'u':
{
unsigned n = va_arg(arg, unsigned);
char lsh = '0' + n % 10;
n /= 10;
char buf[9];
char *p_buf = buf;
while (n) {
*p_buf++ = '0' + n % 10;
n /= 10;
}
while (p_buf != buf)
kputchar(*--p_buf);
kputchar(lsh);
}
break;
case 'o':
{
unsigned n = va_arg(arg, unsigned);
char lsh = '0' + n % 8;
n /= 8;
char buf[10];
char *p_buf = buf;
while (n) {
*p_buf++ = '0' + n % 8;
n /= 8;
}
while (p_buf != buf)
kputchar(*--p_buf);
kputchar(lsh);
}
break;
case 'x':
{
unsigned n = va_arg(arg, unsigned);
char lsh = to_hex_digit(n % 16);
n /= 16;
char buf[7];
char *p_buf = buf;
while (n) {
*p_buf++ = to_hex_digit(n % 16);
n /= 16;
}
while (p_buf != buf)
kputchar(*--p_buf);
kputchar(lsh);
}
break;
case 'X':
{
unsigned n = va_arg(arg, unsigned);
char lsh = to_hex_digit(n % 16);
n /= 16;
char buf[7];
char *p_buf = buf;
while (n) {
*p_buf++ = to_hex_digit(n % 16);
n /= 16;
}
while (p_buf != buf)
kputchar(toupper(*--p_buf));
kputchar(toupper(lsh));
}
break;
case 'c':
kputchar(va_arg(arg, int));
break;
case 's':
kprint(va_arg(arg, char *));
break;
case 'p':
{
kprint("0x");
size_t ptr = va_arg(arg, size_t);
char lsh = to_hex_digit(ptr % 16);
ptr /= 16;
char buf[15];
char *p_buf = buf;
while (ptr) {
*p_buf++ = to_hex_digit(ptr % 16);
ptr /= 16;
}
while (p_buf != buf)
kputchar(*--p_buf);
kputchar(lsh);
}
break;
case '%':
kputchar('%');
break;
default:
kputchar('%');
kputchar(*format);
}
} else
kputchar(*format);
++format;
}
}
void kprintf(const char *format, ...) {
va_list arg;
va_start(arg, format);
kvprintf(format, arg);
va_end(arg);
}
src/kmain.c
#include "uart/uart.h"
#include "syscon/syscon.h"
#include "common/common.h"
#define ARCH "RISC-V"
#define MODE 'M'
void kmain(void) {
uart_init(UART_ADDR);
kprintf("Hello %s World!\n", ARCH);
kprintf("We are in %c-mode!\n", MODE);
poweroff();
}
Running the project
Invoke make run
in the project root. You should see the following output:
Hello RISC-V World! We are in M-mode! Poweroff requested
Final remarks and going further
This is by no means an end to your OSDev adventures on RISC-V. Make the OS kernel your own! Add support for memory management, interrupt handling, porting newlib ... you name it. Give your own RISC-V OS a creative name! "maRVelOS" / marvelos
is already taken by User:Donaldsebleung though ;-)
Also note that this example RISC-V OS runs in M-mode usually reserved for firmware, rather than the S-mode recommended for RISC-V supervisors (OSes). If you wish to follow the RISC-V conventions closely, you may want to look into RISC-V privilege modes and OpenSBI early on and port your OS kernel accordingly. More information about RISC-V privilege modes available on our wiki.