1 Getting Started with Akaros
2 ==========================
5 Last thorough update: **2013-02-07**
11 + From Download to Hello World
12 - Cross compiler (and glibc)
16 - Building and Loading a Virtual Machine Image
20 - Other Dev Workflow Stuff
22 + Other Developer's Stuff
27 In this document, I outline what steps I go through to set up my development
28 environment. Some devs use other setups, and they can put sections farther
29 down in this document describing their steps.
31 Due to the nature of different workflows and different Linux distros, you'll
32 need to do some configuration of your environment. Things won't magically work
38 First off, if you get stuck, email someone. You can join our mailing list by
40 [akaros+subscribe@googlegroups.com](mailto:akaros%2Bsubscribe@googlegroups.com)
41 or visit the [group](http://groups.google.com/group/akaros). Once you've
42 joined, send your messages to <mailto:akaros@googlegroups.com>.
44 Alternatively, you can poke your head in #akaros on `irc.freenode.net`. I'm
45 usually idling in there (alone), and if I'm at my computer, I'll respond.
48 3. From Download to Hello World
49 ----------------------------
50 I'll describe how to get x86 working. RISCV is similar.
52 To start off, make sure AKAROS_ROOT and AKAROS_XCC_ROOT are set in your
53 environment. AKAROS_ROOT is the Akaros repo directory. AKAROS_XCC_ROOT is a
54 directory of your choosing where the toolchain will be installed (more on that
55 in Section 3.1 below).
57 The first step is to configure the kernel. Targets like `config`,
58 `menuconfig`, and some of the other KBuild targets work. Defconfig gives you a
59 default configuration. For example, to config for 64-bit x86:
61 `$ make ARCH=x86 defconfig`
63 Alternatively, you can run menuconfig to customize what settings you want:
65 `$ make ARCH=x86 menuconfig`
67 For x86, you can choose between 32 and 64 bit when you `make menuconfig`. This
68 selection must match your cross compiler `make` command. The default is 64
71 There are a lot of other settings when you `make config`, and you should browse
72 through to decide what you want to enable/disable.
74 Most everyone wants KFS turned on (Filesystems --> KFS filesystem). This is
75 the in-memory filesystem that the kernel uses. The kernel build scripts will
76 look at the "KFS/Initramfs paths" string and take any of those directories and
77 add them to a CPIO archive that will eventually become the root filesystem when
78 Akaros runs. These settings are set by default when you do a `make defconfig`.
80 There are also settings for `ext2`. If you turn on `ext2` support, you need to
81 point to an `img` file that has been formatted with `ext2` and has files in it.
82 If you aren't messing with filesystems at all, feel free to ignore this. It's
83 an in-memory filesystem, like KFS (linked to the end of the kernel), so you
84 won't gain much by using it for now.
86 ### 3.1 Cross compiler (and glibc)
87 The second step is to build the cross compiler, which lives in
88 `tools/compilers/gcc-glibc`
90 `$ cd tools/compilers/gcc-glibc`
92 In this directory, you first need to set up your Makelocal file. There is a
93 template to work from.
95 `$ cp Makelocal.template Makelocal`
97 You need to set your `INSTDIRS` to some place where you want the cross compiler
98 installed. I have a directory named `akaros-gcc-glibc` for this.
100 Additionally, you must set the environment variable `$AKAROS_XCC_ROOT` to point
101 to the installation directory for your architecture. For instance, my
104 `/home/brho/classes/akaros/akaros-gcc-glibc/install-x86_64-ucb-akaros-gcc/`
106 You also need to add `bin` directories to your `PATH` where the cross compiler
107 will be installed. This will vary based on your value for `INSTDIRS`. For
108 instance, my path contains:
110 `/home/brho/classes/akaros/akaros-gcc-glibc/install-x86_64-ucb-akaros-gcc/bin`
112 You can also set up `MAKE_JOBS`, so you don't over or under load your system when
113 building. I have a 2 core laptop, so I use `MAKE_JOBS := 3`
115 At this point, you can build (for example):
119 This might take a while (10-20 minutes for me on a 2007 era laptop).
121 Just to double check everything installed correctly, you should be able to run
122 `x86_64-ucb-akaros-gcc` from your shell.
124 Now, you have a cross compiler ready, and you can start to build Akaros.
127 `cd` back into the repo root.
129 Like the cross compiler, the kernel has its own `Makelocal`.
131 `$ cp Makelocal.template Makelocal`
133 This file is used to set up custom make targets that are not part of the
134 default `Makefile`, but fit nicely into your personal workflow. This file is
135 not under version control and can me made to contain just about anything.
137 Now you're ready to build the kernel:
141 So the kernel built, but you can't do much with it, and you probably have no
144 Notice that we didn't have to set the `ARCH` variable this time. The make
145 system knows what architecture we are set up for and will always build for that
146 architecture until a new `ARCH` is selected (i.e. via `make ARCH=xxx defconfig`
150 First, you'll need to build a few common applications and libraries:
152 `$ make apps-install`
154 Then you can build the tests and small utility programs:
158 You now have programs and libraries, and need to put them in KFS. To do this,
159 we provide a `fill-kfs` make target.
163 The `fill-kfs` target copies your cross compiler's shared libraries and all
164 test binaries into the first "KFS/Initramfs path" you set during configuration
165 (or `kern/kfs/lib` if you just kept the default).
167 Now that you've changed the contents of KFS's source, remake the kernel. You
168 should see something like the following before the kernel links. If you don't
169 see this, then you probably didn't actually fill KFS properly.
173 Adding kern/kfs to initramfs...
178 Busybox provides our shell and some basic utilities. You almost certainly want
179 to build and install busybox.
181 Userspace programs like busybox need to be compiled with the cross compiler and
182 then have their binaries copied to `kern/kfs/bin`. Since most everyone wants
183 busybox and we have a few patches of our own, we have support for automatically
184 building and installing it to KFS.
186 For the default build (`x86_64`):
189 $ cd tools/apps/busybox
190 $ make [x86_64|riscv]
194 And you should be set. Check `kfs` to make sure everything installed. You
195 should get sane results from:
198 $ ls -l kern/kfs/bin | grep cat
199 lrwxrwxrwx 1 brho brho 7 Jan 23 09:19 cat -> busybox
202 You can customize your busybox installation, including the install prefix, the
203 `.config` file, and `make` jobs. Check out the makefile in
204 `tools/apps/busybox` for details.
206 Now that you've changed KFS, don't forget to remake the kernel.
208 ### 3.5 Building and Loading a Virtual Machine Image
209 At this point, you probably have a runnable kernel with programs in KFS. It
210 should be sitting at `obj/kern/akaros-kernel`. When running in a VM, you can
211 either run the kernel directly from `qemu`, or put it in a virtual machine
214 If you don't want to bother with the image, skip this section. I tend to run
215 my images off an image file, since `qemu` acts more like hardware (as far as
216 multiboot goes). The downside is the boot up is slower, especially if you have
217 a large kernel (>100MB). It also takes some effort to set up the VM image.
219 If you are still reading, you'll need an image file that looks like a hard disk
220 to boot `qemu` off of. I put one similar to mine at:
221 <http://akaros.cs.berkeley.edu/files/hdd268mb.img>
223 It's around 268MB (256MiB, or whatever). If you want to make your own, check out
224 [Documentation/howtos/make-bootable-grub-hdd.txt](Documentation/howtos/make-bootable-grub-hdd.txt).
225 That's actually the original document I made back when I first figured it out
226 back in 2009, which was updated again in 2013. In between, I wrote it up
228 <http://www.omninerd.com/articles/Installing_GRUB_on_a_Hard_Disk_Image_File>,
229 which has some other tidbits in the comments. Both methods still use `grub1`.
231 Anyway, I put that img in `AKAROS-ROOT/mnt/`, and make a folder next to it:
232 `AKAROS-ROOT/mnt/hdd`. `mnt/hdd` is the mount point where I mount `hdd.img`
233 (Note I don't call it `hdd64mb.img` on my dev machine).
235 Personally, I always have `hdd.img` mounted. Some of the other devs have make
236 targets that mount and umount it. Whenever I reboot my development machine, I
237 run a script (as root) that mounts the image file and sets up a few things for
238 networking. I put a script I use for this in `scripts/kvm-up.sh`. You'll likely
239 want to copy it to the directory **above** the akaros root directory and edit it
240 accordingly. Feel free to comment out the networking stuff. That's for using
241 networking in `qemu`.
243 Now that your image file is mounted at `mnt/hdd`, you'll want to copy your
244 freshly built kernel to the root of the image. I have a make target in my
245 makelocal for this, so that whenever I do a `make kvm`, it builds the kernel
246 and copies it to my `hdd.img`.
248 I added edited versions of my KVM (and USB) make targets to the
249 `Makelocal.template`. Uncomment the KVM one (at least).
251 Incidentally, I also have the following in my `Makelocal`, so that `make` (and
252 `make all`) also make kvm:
256 Now, `make kvm`. You should be able to see the new kernel in `mnt/hdd/` (do an
257 `ls -l` and check the timestamp).
260 Here is the command I use to run `qemu`/`kvm`. It's evolved over the years,
261 and it will vary based on your linux distribution. Don't run it just yet:
264 $ qemu-system-x86_64 -s -enable-kvm -cpu kvm64 -smp 8 -m 4096 -nographic -monitor /dev/pts/3 -net nic,model=e1000 -net user,hostfwd=tcp::5555-:22 -net dump,file=/tmp/vm.pcap -drive file=mnt/hdd.img,index=0,media=disk,format=raw
267 If you skipped making a virtual machine image or want to run the kernel
268 directly without emulating a disk, replace the "`-drive`" parameter (all the way
269 to `format=raw`) with "`-kernel obj/kern/akaros-kernel`".
271 The `-monitor` is the qemu monitor, which is a CLI for qemu. Pick a
272 tab/terminal/pty in Linux that you will only use for qemu monitoring, and enter
273 `tty'. Whatever it tells you, put in place of `/dev/pts/3`. I've been using
274 the same tab for about 4 years now. In that tab, enter '`sleep 999999999`'.
275 Qemu will still access it, but you won't have to worry about bash trying to
278 `-nographic` allows qemu to work in the terminal you run qemu from, instead of
279 spawning off a fake cpu crt/monitor.
281 The command as written uses qemu's user networking. It's emulated and a little
282 slow. The example I have alo forwards port `5555` on the host to port `22` on
283 the guest. Customize it according to your needs.
285 Another option for networking is to set up a tun/tap device. I use this on
286 some machines, and the kvm-up script has some commands to set it up. It's
287 tricky, and may vary for your distribution. If you do use the tun/tap
288 networking, replace the "`-net user`" section with:
290 `-net tap,ifname=tap0,script=no`
292 The "`-net dump`" option saves a pcap trace of the network traffic. This is very
293 useful for debugging, but probably not needed for most people.
295 Feel free to pick different values for the number of cpus and RAM (8 and 4096
298 Once you finally run it, you can stop the VM by entering '`q`' to the qemu
299 monitor (or just killing the process).. Other help commands from the monitor
300 include '`info cpus`', '`info registers`', '`x`', and '`help`'.
302 In more recent versions of qemu, `CTRL-C` will not get sent to the guest;
303 instead it will kill the VM. If this gets annoying, you can remap "interrupt"
304 to something other than `CTRL-C` in the terminal where you run qemu:
313 ### 3.7 Running on Hardware
314 I have a few bootable USB sticks with grub set up to run Akaros. The make usb
315 target (example in `Makelocal.template`) will copy freshly made kernels to your
316 USB device. You'll need to adjust those paths according to your distro. My
317 usb sticks are usually `/dev/sdc`, for instance (some odd USB device in the
318 last couple of years has taken over `/dev/sdb`. Probably something to do with
319 `udev` changing over the years).
321 Anyway, you'll need to mess around a bit to get that working. Or I can `dd` one
322 for you (I have 4GB disks in my office that I use). If you make your own, the
323 critical part is getting grub to pick the right device (from what I remember),
324 and its fairly similar to installing grub on any old hard drive (since it's
325 just a bloc device). Much easier than a hard disk image file.
329 So now you can run the kernel. It's time to edit a program (or make your own).
330 In this, I'll go through my workflow for making changes.
337 (new version in obj/tests/hello)
343 (rebuilds kernel with the new KFS)
347 (following commands are in Akaros)
348 Shift-G (to get to the kernel monitor)
350 ROS(Core 0)> bb (to run busybox)
353 (Should print your message)
356 ### 3.9 Other Dev Workflow Stuff
357 One thing to note is that while we use dynamic linking for `libc`, `parlib`
358 libraries are statically linked with applications. In fact, nowadays **all**
359 Akaros programs need to be linked againt `parlib` (used to be that single-core
360 processes (SCPs) didn't need it).
362 The makefiles won't notice if you change a file in `parlib` and then remake a
363 binary. So if you edit `user/parlib/uthread.c` for example,
364 `tests/pthread_test` won't get rebuilt. Here's what I do:
367 $ vi user/parlib/uthread.c (make awesome change)
369 $ touch tests/pthread_test.c ; make tests
372 This will force the rebuild of `pthread_test`. Older, untouched binaries (e.g.
373 `block_test`), won't get rebuilt. I actually want this in some cases
374 (different versions of `parlib` when I'm running certain tests). Anyway, just
375 pay attention to what you're building. There's not much output in the console,
376 so you should be able to see what's going on all the time. (unlike when
377 building `glibc`...).
379 Oh, and don't forget to:
383 to make sure you run the new `pthread_test`.
385 Additionally, when switching between 32 and 64 bit x86, `make objclean` before
386 filling KFS. This is the easiest way to make sure you get the appropriate
387 libraries loaded in KFS.
389 Early on as a dev, there are lots of times where you accidentally don't run the
390 right program (or kernel) and won't understand why your change isn't happening.
391 A few `printk("WTF\n")`'s later, you realize you didn't have the `hdd.img`
392 mounted, or you didn't fill KFS, or you didn't relink your binaries, or you
393 forgot to save all files in `vi1 (and not just the current buffer). But after
394 doing a couple `hello worlds`, you're set.
396 Alternatively, you could have a make target to run qemu, which also touches all
397 binaries (or otherwise enforces a rebuild), auto-fills KFS, remakes the kernel,
398 and mounts/copies/unmounts your `hdd.img`.
399 Personally, I like to keep track of what is going on under the hood, esp if I
400 want to do something a little differently (like with testing `ext2`, having
401 different versions of `parlib` with some binaries, or being picky about my
406 ---------------------
409 For now, you need a 64 bit distro to build the RISCV stuff, so I don't do it
410 very often. I'll eventually sync up with Andrew and we'll get this part sorted
414 5. Other Developer's Stuff
415 --------------------------