This is cleaner as it removes yet another .gitignore rule and allows
for one command per arch.
This was prompted by the recent Docker patch creating files as root,
which then breaks future host runs.
Also separate all arguments by newlines. This:
- makes it much easier to minimize when reporting QEMU bugs
- shows the full command on the screen despite line wrapping being turned off
by one of the annoying emulators, to be investigated.
The commands are not yet pristine, because:
- there are some leading spaces
- options not sorted
but it is already a huge improvement, and those operations are now be easy from Vim.
The general strategy is:
- end every variable containing a command with a newline
- when expanding such variables, don't add an extra newline
gem5 cache automate benchmarks, and remove results that were scattered all over.
Move console fun under linux kernel section, since not important enough
for getting started.
Set arch x86_64 by default through common on all scripts.
Document that X11 mouse is not moving.
Convert x='' to x= on all scripts, and also fix case ;; indentations.
Add dummy value to QEMU's -trace enable= to prevent warning.
Expand built-in package choice rationale.
Most of it was present inside buildroot/output.* and the rest scattered
on top level.
This came about for the n-th time when we were reviewing QEMU trace file
locations.
On one hand, it would be cool to have per arch traces.
This made buildroot/output.${arch}~/ a natural choice.
But on the other, those traces have nothing to do with Buildroot,
and could potentially interfere with Buildroot build files.
It also feels nicer to have buildroot/ pristine source code only,
and keep all output under a single directory out/
Otherwise it was getting messy to distinguish between trace-events
and trace-<pid> in the .gitignore.
Now only the latest trace will be kept, with a fixed name.
We have wanted to do this since forever, but the last straw was gem5 aarch64,
which runs too fast, and makes it very hard to write "root" withing 60 seconds!
Another possibility to solve that would have been to find an answer for:
https://unix.stackexchange.com/questions/340333/how-can-i-get-bin-login-to-not-timeout
which we didn't, but not typing root at all is even better.
Take this opportunity to document how to login into user0.
sshd: automate and document further
gdbserver: automatically startup internet on /gdbserver.sh
In order to make the system easier to understand.
These include:
* networking
* klogd and syslogd. TODO what are those for? I could not see anything useful that they do.
Also get rid of the useless S20random thing while we are at it.
This lead to an infinite chaing of refactorings:
Fix arm and aarch64 lx-symbols
This was just by chance, a continue GDB command was needed before running it.
I noticed that there is no more need for the x86 GDB --ex 'set arch i386:x86-64:intel'
thing and unified all archs. TODO bisect where that was fixed and understand why.
Add debug kernel configs for gem5 kernel.
Document gem5 aarch64 gdb debug doesn't work.
Document tmux split pane action.
This is necessary because qemu and gem5 now use the same build folder.
Separate ./run -e and -f for kernel options before and after the ' - '.
This was already the better thing to do when -E was introduced,
but lkmc_nonet prompted me to do it nicer now.
Use the common script to find the out_dir on every toplevel script.
Include usage man pages on README.
Don't pass -e on checkpoint restore.
Add benchmarks to how much GEM5 is slower than QEMU.
Rename Kernel boot command line arguments to match kernel docs name.
Document how to pass extra options to GEM5.
Start listing interesting benchmarks to run on GEM5.
Add an openmp hello world.
Semi automated conversion, in part because Pandoc is kind of buggy for adoc:
Some of the commands were along the lines:
for f in *.md; do pandoc --atx-headers --base-header-level=3 -o ${f%.md}.adoc --wrap=none $f; done
sed -Ei '/\[\[.*\]\]/d' *.adoc
while read -r f; do cat $f; echo; done <f >g
