Humble Beginnings
My previous project, relics,
was a massive failure. That self-proclaimed "exokernel" couldn't even get
into userland! On a good day, it would kernel panic if built with O2 (it failed to
build without optimizations -- a sign of bad software). On a bad day, you would
chase down a bug for eleven months. Yeah. I spent 11 months trying to
figure out why interrupts wouldn't work. The bug was, of course, a single
line of code. This project wasn't even in the stages where it could be able
to be used with a debugger, so any debugging had to happen with printf.
What's worse is that CPUs execute very fast and sometimes out-of-order.
This sometimes results in a triple fault and reset in qemu before the code
before it could execute debug printfs. Code like printf("Foo"); triple_fault();
might not even print "Foo" at all! Talk about a debugging nightmare.
One night, out of frustration, I gave up. I decided that I like osdev,
but I don't like the OS I made from scratch. It didn't do nearly what I
wanted and couldn't ever be made to do so. So I took to the internet in
search of easy-to-learn educational operating systems, and someone
suggested xv6: a small operating system made by the MIT PDOS group that
tried to be a small UNIX-like system, while still being teachable.
Armed with this knowledge, I looked through the code and determined it
to be good if I could make it work.
Oh man, I did not know what I would get myself into with this project.
At the time of forking, xv6 had been effectively dead for 8-18 years,
depending on who you ask. The final nail was put in the coffin when they
put a message in the README saying that the project was no longer maintained.
When trying to build this outdated project, gcc will throw a million errors
at you, talking about one for infinite recursion in sh.c, another for some
old-style C, and so forth. Once you finally get the thing to build, it
doesn't run! Luckily,
some savior in the PRs
commented a fix, and I was back to business. After I got it to build, I
called it good and pushed my fork to my own branch, relics, named
after my previous operating system attempt. I ran clang-format on
the source with the style of the linux kernel -- my favorite style.
After that, I separated the .c files for the kernel space and userspace
into directories and added some Makefiles in those directories to better
separate them. This would later prove integral as userspace quickly
diverged, needing many different CFLAGS from the kernel. If you look
in kernel/boot/, you will find initcode64.S. This is a 64-bit port of
initcode.S found in the original build of xv6. Its relevance is that
it is technically userspace code and has no business being in the
kernel directory (much less kernel/boot), but I have gotten used to
it there, so I don't care to move it. It's not like userspace doesn't
use assembly because it does: usys.S.
Anyway, I should wrap this up. This blog is only supposed to go over
the initial history of xv6_64, specifically the first day.
All of this work was done in one day! I will do this for every blogpost,
but the relevant commits are at the bottom of this post for proof.
commit 767541ca11cbec02d42a7a7cd52023188196c7a8 Author: Connor Date: Sun May 26 16:42:48 2024 -0500 Clang fixes commit e5e5004333b352556e3cb7f9b8ed3c2e0c875d49 Author: Connor Date: Sat May 25 22:05:37 2024 -0500 gitignore oops commit b143cd269ac42f7c7d5d8ffff8fdca5496775e30 Author: Connor Date: Sat May 25 22:02:43 2024 -0500 Organize folders more commit 70377e7da44c40c1495c0070680d8e98eac1b5b4 Author: Connor Date: Sat May 25 02:19:55 2024 -0500 Clang-format update commit adb31a31455d08b59f99bf36039548c8c58577e4 Author: Connor Date: Sat May 25 02:09:58 2024 -0500 Initial fork commit eeb7b415dbcb12cc362d0783e41c3d1f44066b17 (origin/master, origin/HEAD) Author: Frans Kaashoek Date: Mon Aug 10 14:28:02 2020 -0400 Be more explicit that we are not maintaining the x86 version anymore