7 ms·

Writing a RISC-V Emulator from Scratch

130 points by vcsilva 6y ago

burakemir 6y ago
This looks like a nice project, and I will check on its progress. So far only two of the advertised steps are there. Reminds me, I started a RISC emulator project on a holiday - and then forgot about it when the holiday was over :o Not sure if combining "learn rust" and "learn RISC V" at the same time is a good combination for a large audience, it will depend on the reader. Maybe the intended audience is someone who knows enough basics and has plenty of motivation for/prior knowledge of/interest in both topics. And then, for good content, there will always be an audience.
- p0llard 6y ago
  > Not sure if combining "learn rust" and "learn RISC V" at the same time is a good combination for a large audience, it will depend on the reader. Heh, this is _exactly_ the combination that appeals to me, or at least would have before I learned either of them.
saagarjha 6y ago
Unfortunately it seems like the privileged instruction set is not yet covered. Also, I think that the arithmetic needs to have explicit wrapping? Otherwise it’ll overflow and cause a trap in debug builds I believe.
a-priori 6y ago
It looks like it's the season for hobby RISC-V emulators, because I've also been working on one. https://github.com/michaelmelanson/riscy https://github.com/michaelmelanson/riscy The tricky part here comes when you have to run non-trivial programs. Mine currently passes the basic test suites, but it still fails in the C runtime startup when executing programs compiled with GCC. My next step is to do co-simulation with another emulator in order to find out where their execution states diverge.
- thethirdone 6y ago
  > Mine currently passes the basic test suites The tests in riscv-tests are not complete enough to indicate strong compliance. I have found several bugs in my implementation that were not caught by them. > but it still fails in the C runtime startup when executing programs compiled with GCC The biggest issue I have seen with running C programs is that people don't have the system calls that C runs on startup. In a log from rv8, you need brk, uname, readlinkat, and mprotect. So without those you shouldn't be able to get a program into user code unless you make a custom entry point. Source: I am the maintainer of RARS (https://github.com/TheThirdOne/rars https://github.com/TheThirdOne/rars)
  - a-priori 6y ago
    > The tests in riscv-tests are not complete enough to indicate strong compliance. I have found several bugs in my implementation that were not caught by them. I've noticed that, yes. They got me pretty far, but there's something I'm doing wrong and I haven't been able to find what. That's why the next thing I'm going to try is co-simulation. > The biggest issue I have seen with running C programs is that people don't have the system calls that C runs on startup. In a log from rv8, you need brk, uname, readlinkat, and mprotect. So without those you shouldn't be able to get a program into user code unless you make a custom entry point. Yes, that hasn't been a problem so far. I've written minimal implementations of a few (so far: fstat, brk, exit) as I've gone along. https://github.com/michaelmelanson/riscy/blob/master/emulator/src/subsystem/posix.rs https://github.com/michaelmelanson/riscy/blob/master/emulato...
forty 6y ago
> This is the book for writing a 64-bit RISC-V emulator from scratch in Rust. Next I want to see the book for writing a 64-bit RISC-V emulator from rust in Scratch ;) The upper case S in the title made me believe for a second that they did it in Scratch :)
- CDSlice 6y ago
  Since I got into programming with Scratch, every time I see it capitalized in a HN title I get my hopes up that someone made something really cool in it. Alas, it never happens.
  - shakna 6y ago
    This [0] project might be one of my favourite ones I've seen in Scratch. I don't know how they managed to correctly keep everything together, considering the size of some of the internals. But it does do almost everything I would expect. [0] https://scratch.mit.edu/projects/159487877/ https://scratch.mit.edu/projects/159487877/
azhenley 6y ago
Making an OS for RISC-V in Rust: http://osblog.stephenmarz.com/ http://osblog.stephenmarz.com/
d0iasm 6y ago
Thanks for your interest! I'm an author of this book. This book is very beginning as you mentioned and I'm still working on it. I already succeeded to run xv6 in my RISC-V emulator and wrote a blog post "Made a RISC-V Emulator Running Xv6": https://d0iasm.github.io/blog/risc-v/2020/04/03/xv6-on-my-riscv-emulator.html https://d0iasm.github.io/blog/risc-v/2020/04/03/xv6-on-my-ri... I have 2 different RISC-V emulators and its goals are different: rvemu (https://github.com/d0iasm/rvemu https://github.com/d0iasm/rvemu): Can run xv6. It's trying to support Linux and make it faster in the future. rvemu-for-book (https://github.com/d0iasm/rvemu-for-book https://github.com/d0iasm/rvemu-for-book): Reference implementation for the book. Simpler than original one.
- seusscat 6y ago
  Could you please elaborate on what the differences between the two implementations are?
  - d0iasm 6y ago
    The main difference between them is the number of instructions are supported. Rvemu supports RV64GC, while rvemu-for-book will support RV64I and instructions that xv6 uses (e.g. mul). Also rvemu will support a device tree but rvemu-for-book won't.
- mrlonglong 6y ago
  What does opcode 0x41 does? It's stopping the emulator when I try to run it on fib.text. Otherwise it's a pretty decent tutorial. I've also been following Stephen Marz's tutorial on bringing up a basic operating system on RiscV 64, going to use the two together.
snvzz 6y ago
This will be interesting once it is finished. As it is now, it is just a skeleton, and thus not worth the light of day.

130 points by vcsilva 6y ago

This looks like a nice project, and I will check on its progress. So far only two of the advertised steps are there. Reminds me, I started a RISC emulator project on a holiday - and then forgot about it when the holiday was over :o Not sure if combining "learn rust" and "learn RISC V" at the same time is a good combination for a large audience, it will depend on the reader. Maybe the intended audience is someone who knows enough basics and has plenty of motivation for/prior knowledge of/interest in both topics. And then, for good content, there will always be an audience.

p0llard 6y ago
> Not sure if combining "learn rust" and "learn RISC V" at the same time is a good combination for a large audience, it will depend on the reader. Heh, this is _exactly_ the combination that appeals to me, or at least would have before I learned either of them.

saagarjha 6y ago

Unfortunately it seems like the privileged instruction set is not yet covered. Also, I think that the arithmetic needs to have explicit wrapping? Otherwise it’ll overflow and cause a trap in debug builds I believe.

a-priori 6y ago

It looks like it's the season for hobby RISC-V emulators, because I've also been working on one. https://github.com/michaelmelanson/riscy https://github.com/michaelmelanson/riscy The tricky part here comes when you have to run non-trivial programs. Mine currently passes the basic test suites, but it still fails in the C runtime startup when executing programs compiled with GCC. My next step is to do co-simulation with another emulator in order to find out where their execution states diverge.

thethirdone 6y ago
> Mine currently passes the basic test suites The tests in riscv-tests are not complete enough to indicate strong compliance. I have found several bugs in my implementation that were not caught by them. > but it still fails in the C runtime startup when executing programs compiled with GCC The biggest issue I have seen with running C programs is that people don't have the system calls that C runs on startup. In a log from rv8, you need brk, uname, readlinkat, and mprotect. So without those you shouldn't be able to get a program into user code unless you make a custom entry point. Source: I am the maintainer of RARS (https://github.com/TheThirdOne/rars https://github.com/TheThirdOne/rars)
- a-priori 6y ago
  > The tests in riscv-tests are not complete enough to indicate strong compliance. I have found several bugs in my implementation that were not caught by them. I've noticed that, yes. They got me pretty far, but there's something I'm doing wrong and I haven't been able to find what. That's why the next thing I'm going to try is co-simulation. > The biggest issue I have seen with running C programs is that people don't have the system calls that C runs on startup. In a log from rv8, you need brk, uname, readlinkat, and mprotect. So without those you shouldn't be able to get a program into user code unless you make a custom entry point. Yes, that hasn't been a problem so far. I've written minimal implementations of a few (so far: fstat, brk, exit) as I've gone along. https://github.com/michaelmelanson/riscy/blob/master/emulator/src/subsystem/posix.rs https://github.com/michaelmelanson/riscy/blob/master/emulato...

forty 6y ago

> This is the book for writing a 64-bit RISC-V emulator from scratch in Rust. Next I want to see the book for writing a 64-bit RISC-V emulator from rust in Scratch ;) The upper case S in the title made me believe for a second that they did it in Scratch :)

CDSlice 6y ago
Since I got into programming with Scratch, every time I see it capitalized in a HN title I get my hopes up that someone made something really cool in it. Alas, it never happens.
- shakna 6y ago
  This [0] project might be one of my favourite ones I've seen in Scratch. I don't know how they managed to correctly keep everything together, considering the size of some of the internals. But it does do almost everything I would expect. [0] https://scratch.mit.edu/projects/159487877/ https://scratch.mit.edu/projects/159487877/

azhenley 6y ago

Making an OS for RISC-V in Rust: http://osblog.stephenmarz.com/ http://osblog.stephenmarz.com/

d0iasm 6y ago

Thanks for your interest! I'm an author of this book. This book is very beginning as you mentioned and I'm still working on it. I already succeeded to run xv6 in my RISC-V emulator and wrote a blog post "Made a RISC-V Emulator Running Xv6": https://d0iasm.github.io/blog/risc-v/2020/04/03/xv6-on-my-riscv-emulator.html https://d0iasm.github.io/blog/risc-v/2020/04/03/xv6-on-my-ri... I have 2 different RISC-V emulators and its goals are different: rvemu (https://github.com/d0iasm/rvemu https://github.com/d0iasm/rvemu): Can run xv6. It's trying to support Linux and make it faster in the future. rvemu-for-book (https://github.com/d0iasm/rvemu-for-book https://github.com/d0iasm/rvemu-for-book): Reference implementation for the book. Simpler than original one.

seusscat 6y ago
Could you please elaborate on what the differences between the two implementations are?
- d0iasm 6y ago
  The main difference between them is the number of instructions are supported. Rvemu supports RV64GC, while rvemu-for-book will support RV64I and instructions that xv6 uses (e.g. mul). Also rvemu will support a device tree but rvemu-for-book won't.
mrlonglong 6y ago
What does opcode 0x41 does? It's stopping the emulator when I try to run it on fib.text. Otherwise it's a pretty decent tutorial. I've also been following Stephen Marz's tutorial on bringing up a basic operating system on RiscV 64, going to use the two together.

snvzz 6y ago

This will be interesting once it is finished. As it is now, it is just a skeleton, and thus not worth the light of day.