fwsgonzo / barebones Goto Github PK

I have added support for ubsan now, however there is a problem with using clang, as its referencing some C++ typeid structures. I tried adding -fno-sanitize=vptr, but it had no effect on linking.

hello,

Get this error when trying to build_iso.sh on MacOS

-- Configuring done
-- Generating done
-- Build files have been written to: /Users/thierry/Dev/barebones/machines/default/build
Scanning dependencies of target exceptions
Scanning dependencies of target tinyprintf
[ 7%] Building C object ext/CMakeFiles/tinyprintf.dir/tinyprintf/tinyprintf.c.o
[ 7%] Creating directories for 'exceptions'
[ 11%] Performing download step (download, verify and extract) for 'exceptions'
-- Downloading...
dst='/Users/thierry/Dev/barebones/machines/default/build/exceptions/src/exceptions.zip'
timeout='none'
-- Using src='https://github.com/fwsGonzo/barebones/releases/download/exceptions/exceptions.zip'
[ 14%] Linking C static library libtinyprintf.a
[ 14%] Built target tinyprintf
Scanning dependencies of target kernel
[ 25%] Building C object src/CMakeFiles/kernel.dir/hw/serial1.c.o
[ 25%] Building C object src/CMakeFiles/kernel.dir/crt/c_abi.c.o
[ 25%] Building C object src/CMakeFiles/kernel.dir/kernel/kernel_start.c.o
/Users/thierry/Dev/barebones/src/hw/serial1.c:4:48: error: argument to 'section' attribute is not valid for this target: mach-o section specifier requires a segment and
section separated by a comma
static char initialized attribute((section(".data"))) = 0;
^
1 error generated.
make[2]: *** [src/CMakeFiles/kernel.dir/hw/serial1.c.o] Error 1
make[2]: *** Waiting for unfinished jobs....
/Users/thierry/Dev/barebones/src/kernel/kernel_start.c:6:13: warning: unused function '__init_paging' [-Wunused-function]
static void __init_paging()
^
1 warning generated.

A glaring flaw is the lack of a 512b minimal bootloader that ATA loads the kernel to its dest and then enters protected and jumps to _start. One reason to add one is the extra swag of having very very tiny kernels that still can do stuff.

Several problems:

1. Theres not enough room to really inspect ELFs and validate stuff, GRUB will at least tell you if something is wrong even if its a big blob. So, either the bootloader just jumps to <some offset in the ELF that corresponds to the 64-bit program entry>, or we need to burn the _start location into the bootloader itself via a script.
2. We no longer get memory maps from GRUB so we have to do without. CMOS chip will tell us about up to 64mb of memory, which is basically nothing today. It really is a problem.
3. It will likely be brittle on real hardware - especially the ATA loader code. It also won't work on modern clouds like GCE which demands you boot using virtio-scsi.

In short, it may just not be worth it.

I can't find any mention of the license in the repository, which one is it? Is it public domain or something more restrictive?

There is a case for bundling libcxxabi with the project so that we can potentially enable RTTI and exceptions in C++. This will come at a cost in binary size, for the most part, but it will at the very least be a CMake option. Sadly, the archive may have to be compiled beforehand and bundled as a download.

The archive will likely bring in a few extra dependencies on C standard functions.

There is clearly a problem with gc-sections and 32-bit builds. Does not enter kernel_start properly, so it's a linker, alignment or both issue.

How did you compile libc++abi.a and libgcc.a?