mission ------- Build a small Linux system with as little resource usage as possible. Modern kernel, suckless if possible, i486 as target. Trying to squeeze binaries with NetBSD cruch. Using a floppy boot loader which uses a USTAR-like format for storing kernel and ramdisk. Using musl for its ability to produce small static binaries. requirements ------------ - A native C compiler. gcc/clang is a must for the Linux kernel - NASM for assembling the floppy boot loader 'uflbbl' - a make, probably GNU make is required directory layout ---------------- scripts/ build scripts lib/ libraries for build scripts dowloads/ downloaded software (tarballs or git/svn/cvs checkouts) src/ contains all unpacked sources for building ../stage0 host toolchain building ../stage1 ramdisk first native build configs/ configurations used during building, e.g. suckless config.h files build/ the binary artifacts of each stage ../stage0 ../stage1 local/ local adaptions and configurations which go into the ramdisk ramdisk/ the ramdisk being later compressed to ramdisk.img root/ the root filesystem later being used via NBD or directly as image tests/ test programs to test sanity of stages tools/ various tools needed on the host to build the distro patches ------- - tcc * musl uses 't' and 'u' for i487 stack inline assembly, we try not to use floating point anywhere, currently the tcc patch is a non-functional stub to make musl compile * configure fails completly when used with the stage0 compiler, we patch by hand 'configure' and 'Makefile' to build a standalone version using stage0 libaries and header files * 'config.mak' sets correct pathes and CPU flags for the i386 architecure - musl * tcc doesn't expand macro definitions in inline assembly: SYSCALL_INSNS in syscall stubs. Expanded them by hand. * tcc has no support for 'Complex', we just define it empty and we never use complex numbers * replaced 'weak hidden' with '__attribute__((weak)) __attribute__((hidden))' which is more standard conform * 'fabs', 'fmod' have assertions to not be used, they use inline assmmbly for i487 stack inline assembly markers * '*rint': fistpll to flds, tcc has not everywhere the same opcode mnemonics as gcc and clang * 'sigsetjmp': replaced 'jecxz 1f' with 'jecxz 1b' and converting the far forward jump to a predefied near jump - ubase * 'sys/sysmacros.h' is missing for 'minor' and 'major' - smdev * 'sys/sysmacros.h' is missing for 'minor' and 'major' - netbsd-curses * ignoring 'strong alias' attribute - vis * disabled probing of 'curses/termcap' with 'pkconfig', added them directly to 'configure' - dropbear * patched binary locations and pathes in default_options.h - tinyxlib * added a missing stdint.h * patched compiler and linker flags - tinyxserver * patched installation * patched compiler and linker flags - xauth * disabled IPv6 - xhost * disabled IPv6 * fixed some missing IPv6 conditionals * added conditional for FamilyServerInterpreted ("si:") - image viewer - meh * disabled external libraries for gif, png, jpeg, imagemagick * fixed dependency generator - feh as alternative? - slock * fixed for missing setgroups in musl * disabled XRandr extension - lua * removed dependency on readline - wordgrinder * TODO: work in progress - rxvt * use the 'fixed' font per default and only that one - notion * disabled XCB for keysum definitions * patched for using the stage1 libraries and header files * heavily patched Xrandr so at least the lua module builds (it cannot possibly work though), made the module return some default VGA mode values to at least work as a dummy. * linker fix for -rdynamic * patches for building statically and doing a lua module preload * disabled Xinerama dependencies ------------ - vis - libevent - libtermcap - netbsd-curses - optionally lua - tmux - libevent - netbsd-curses (compat.h:379: error: incompatible types for redefinition of 'forkpty', some non-compiling fallback when there is no curses?) - mandoc - zlib - lua51 - tinyxserver - tinyxlib - some basic fonts for default and cursor kernel options -------------- We just enable things we absolutely need to be running Linux, each config option below states the main reason we think, it has to be enabled. The starting point of the Linux kernel configuration is 'make tinyconfig'. - CONFIG_M486SX=y - go to the lowest supported kernel "This kernel requires an i686 CPU" - CONFIG_TTY=y - we need some TTY support for any kind of output - CONFIG_VT_CONSOLE=y CONFIG_VGA_CONSOLE=y - usual VGA virtual consoles - CONFIG_SERIAL_8250=Y CONFIG_SERIAL_CORE_CONSOLE=Y - serial console without kernel output - CONFIG_PRINTK=y CONFIG_X86_VERBOSE_BOOTUP=y CONFIG_EARLY_PRINTK=y CONFIG_BUG=y CONFIG_DEBUG_BUGVERBOSE=y CONFIG_FRAME_POINTER=y - enable an early boot console (otherwise we don't see any errors anymore) "Booting from ROM... "early console in setup code" - show early boot messages and early console messages - enable bug reporting of the kernel - frame pointers allow stack strace in case of errors - CONFIG_MAGIC_SYSRQ=y - for debugging - CONFIG_LOG_BUF_SHIFT=16 CONFIG_PRINTK_SAFE_LOG_BUF_SHIFT=12 - trying to make some buffers smaller, the effect might not be that big - CONFIG_BLK_DEV_INITRD=y CONFIG_RD_XZ=y - allow an XZ compressed ramdisk - CONFIG_ELF_CORE=y CONFIG_BINFMT_ELF=y - "Run /bin/sinit as init process Failed to execute /bin/sinit (error -2)" - MODIFY_LDT_SYSCALL - "Run /bin/sinit as init process process '/bin/sinit' started with executable stack traps: sinit[1] general protection fault ip:804b0e8 sp:bfa84c10 error:0 in sinit[8049000+5000]" => this is the error we get - musl (https://stackoverflow.com/questions/30377020/on-linux-is-tls-set-up-by-the-kernel-or-by-libc-or-other-language-runtime) "I haven't looked at glibc yet, but I just checked musl. It initializes the TLS on startup (as you said), but then also uses a set_thread_area syscall to make an entry in the process' LDT which points to the space allocated for TLS. Then libc sets the %gs register so that it refers to the LDT entry for TLS. (This is on x86-32 -- x86-64 uses a different segment register.)" – – Alex D May 28, 2015 at 6:48 - CONFIG_POSIX_TIMERS=y - "Run /bin/sinit as init process process '/bin/sinit' started with executable stack process 1 (sinit) attempted a POSIX timer syscall while CONFIG_POSIX_TIMERS is not set" - CONFIG_BINFMT_SCRIPT=y - "Run /bin/sinit as init process ... "execvp: Exec format error" => we try to execute a shell script - CONFIG_DEVTMPFS=y DEVTMPFS_MOUNT=y - one visible effect is that getty cannot spawn a login session (and on the serial console you see tons of errors about /dev/ttyX not being available) - TODO: do we still need that if we have smdev? - CONFIG_PROC_FS=y - "login: ttyname: No such file or directory" - CONFIG_MULTIUSER=y CONFIG_NET=y - "login: initgroups: Function not implemented" - musl (in getpwnam) - nscd query code which calls 'socket' and then only falls back to using files, this is a problem in musl. - CONFIG_PROC_SYSCTL=y (CONFIG_PROC_PAGE_MONITOR=y ?) CONFIG_SYSFS=y => we want to have /proc and /sys => not quite sure, but either CONFIG_PROC_FS or CONFIG_SYSFS is needed) - CONFIG_UNIX98_PTYS "mount: mount: devpts: No such device" - CONFIG_COMPAT_32BIT_TIME - for 32-bit time parameter as in pselect (this of course made the whole thing non-2038-safe) "vis: mainloop function not implemented", needs 'pselect' "pselect, function not implemented SYS_pselect6_time64" - CONFIG_MSDOS_PARTITION - for fdisk - CONFIG_MODULES - modules support, for everything which doesn't has to be in the kernel, we can have a module and load it in rc.init. - CONFIG_UNIX (networking options, as module), unix domain sockets - tmux - CONFIG_PCI_QUIRKS=y "8139cp: 8139cp: 10/100 PCI Ethernet driver v1.3 (Mar 22, 2004) 8139cp 0000:00:03.0: found PCI INT A -> IRQ 11 8139cp: probe of 0000:00:03.0 failed with error -22" 22 is EINVAL, this is a nasty one to find - CONFIG_VGA_ARB => might be needed for PCI not to choke on VGA cards - CONFIG_BLK_DEV_FD=m => we might want to use floppies for confugration or files - CONFIG_BLK_DEV_NBD=m => NBD support for root filesystem on network block device - CONFIG_EXT4_FS=m, CONFIG_EXT4_USE_FOR_EXT2=y => mount ext4 filesystems (for instance the root.img) Debugging: - CONFIG_KALLSYMS=y CONFIG_KGDB=y CONFIG_UNWINDER_FRAME_POINTER=y CONFIG_DEBUG_INFO_DWARF_TOOLCHAIN_DEFAULT=y CONFIG_FRAME_POINTER=y CONFIG_UNWINDER_FRAME_POINTER=y Memory optimization: - CONFIG_PHYSICAL_START=0x1000000 16M CONFIG_PHYSICAL_ALIGN=0x2000 8K CONFIG_RELOCATABLE=n boot.asm: set ramdisk and linux addresses correctly tinyx/framebuffer: - CONFIG_FB=y => generic framebuffer support - CONFIG_FRAMEBUFFER_CONSOLE=y => console on framebuffer - CONFIG_FRAMEBUFFER_CONSOLE_LEGACY_ACCELERATION=y - CONFIG_FRAMEBUFFER_CONSOLE_DETECT_PRIMARY=y - CONFIG_FONTS => do we need them, at least fk.ko complains about font functions otherwise - CONFIG_FB_VESA=y => only selectable if framebuffer support is NOT built as modules => this also means we can only start x framebuffer X in the most generic SVGA case when using no modules for the framebuffer modules => works only if a vga=XXX parameter is passed on boot? TODO: CONFIG_FB_VGA16=m => for lowest VGA mode => never got that one working => not sure if that one is needed when choosing VGA in framebuffer TODO: looks optional for a simple running system - CONFIG_FRAMEBUFFER_CONSOLE_LEGACY_ACCELERATION=y => we use fbdev low-level drivers for Cirrus, Alix, etc. - CONFIG_FB_CIRRUS=y => for the Unisys 486 machine, also for qemu? - CONFIG_FB_GEODE_LX=y => for the Alix 1.E TODO: needed? and for which cards? - CONFIG_FB_SIMPLE=y - CONFIG_FB_DEVICE=y => to get a /dev/fb* => TODO: check if XF86fbdev need that TODO: which one of the hardware acceleration options are needed for which cards and have performace benefits (Cirrus for instance should have BtBlt functions) - CONFIG_FB_TILEBLITTING=y TODO: what happens when we have multiple framebuffer drivers, which one takes precedence (for instance SVGA over VGA16 over CIRRUS?). - CONFIG_MOUSE_PS2=m - CONFIG_INPUT_MOUSEDEV=m CONFIG_INPUT_MOUSEDEV_PSAUX=y - CONFIG_INPUT_EVDEV=m => /dev/input/eventX - CONFIG_SYSFB_SIMPLEFB=y => so we get output after the vga= mode is set, so the output continues on the VGA text console TODO: CONFIG_FRAMEBUFFER_CONSOLE_DETECT_PRIMARY=y TODO: FRAMEBUFFER_CONSOLE_DEFERRED_TAKEOVER=y TODO FROM HERE: TODO: preemption better for 1 CPU systems, Desktop? BLOCK_LEGACY_AUTOLOAD?? needed for /dev/ram0? are we doing pivot_root ? - CONFIG_LEGACY_PTYS=y CONFIG_LEGACY_PTY_COUNT=16 - CONFIG_TIMERFD - for sdhcp cd ../linux #TODO [*] Multiple users, groups and capabilities support [*] Legacy autoloading support [*] Enable core dump support [*] RAM block device support [*] AT keyboard (NEW) [*] PS/2 mouse (NEW) bugs ---- - getpass in login needs legacy BSD ttys? - tcc cannot compile the linux kernel. Most likely this will remain a bug forever.. so we must use gcc toolchain for building the kernel, tcc doesn't work since a long time (since tccboot) or only in non-public experimental tcc/linux branches - Having i486tcc-linux self-hosting might not be feasable as we have too many funny build tools around and we don't want to get into the busybox/tinderbox trap, where everything has to be emulated. We rather reimplement software to use less build tools (at most tcc and samurai and maybe some confiure-like new tool). - 'joe' is quite slow in startup. For now we resort to 'vis' with somewhat better performance. It's not clear whether the software is so slow or because we use 'tcc'. - build only the i386 (or choosen cross-compiler), especially the native one for stage1. The stage0 can be a full or a partial cross-compiler ideally. For now both stages are building a full cross-compiler and then superflous architectures are removed. Fixing this likely needs some rewrite of the tcc Makefiles. - The native compiler in stage1 doesn't set pathes correctly: # tcc -o test test.c tcc: error: file 'crt1.o' not found tcc: error: file 'crti.o' not found test.c:1: error: include file 'stdio.h' not found Does recompiling it help? It's just search pathes presumably still being hard-coded to root/stage1, this works for now (as workaround): i386-tcc -vv -static -L/lib -I/include -nostdlib -c -o test.o test.c i386-tcc -vv -static -L/lib -I/include -nostdlib -o test /lib/crt1.o test.o /lib/libc.a so recompiling it inside the chroot might work (though it needs patching again and a stage2 probably, as it has to run inside the chroot and eventually even inside a qemu emulation). - currently we bundle too many things in the initial ramdisk: - all kernel modules (instead of just the essential or/and the ones fitting to the hardware) - kernel modules should remain uncompressed when added to the ramdisk and compressed as *.ko.xz when living in a normal filesystem on a floppy or hard disk. - binaries are not crunched. sbase/ubase/abase create Xbase-box crunched binaries by hand, doesn't work across packages. NetBSD crunch approach could help. - vis has no lua interpreter support build-in currently (needed?) - gave up on ex, is not really working: memory management issues, always results in "Out of memory - too many lines in file" This editor is a little bit too low level to use anyway. - a lot of stuff is copy-pasted between sbase/ubase/abase/smdev, etc. for instance parts of libutil - we heavily use tools from the host (perl, python, make, m4, etc.) from the host (but we don't want those tools inside our system also). So self-hosting will be tricky. - mandoc needs zlib unconditionally, if we don't have compressed man pages, we need to patch this and add a NO_COMPRESSION flag to 'configure.local'. - a poweroff sending SIGUSR1 to sinit would be nice, also shutdown fails by leaving users logged in (stuck in shell, getty is killed, but there is no message on the screen or so) and it doesn't unmount /. (umount2 / error), there is also no 'wall' - minimal binary seed: not that important that we have a small enough requirement list to start with, currently: - host C compiler - host C library to link against first tcc - tons of tools which are not really binary seed (they are just used to produce binaries) - can we use ccache and distcc with tcc? - nbd-client gets killed before calling killall5, we cannot mark it to be started by 'init' presumably (as systemd does), we can also not umount it otherwise we don't have killall5. There is the option of having a shutdown ramdisk (as others do).. - maybe we need a shutdown ramdisk (generated on the fly or using pivot_root and not switch_root and reuse the init ramdisk - but loosing the memory while running and pivot_root is a deprecated syscall. And we have a new ramarchive initrd and not a traditional ramdisk. The shutdown ramdisk is not so much a ramdisk but a tmpfs you are not unmounting with the rest of the file systems. And you populate it with killall5 and unmount, etc. - qemu: Slirp: Failed to send packet, ret: -1 => seems to be a qemu network thingy.. - vis sometimes looses first line of text or shows strange artefacts when exiting it - If building on x86_64, you need lib32 glibc and gcc-libs for tcc to link a standalone version between stage0 and stage1 - does the ramdisk get correctly unmounted? I see a /mnt busy message when doing the switch_root - man pages are not compressed on the root partition, mandoc can deal with compressed pages (as it requires zlib) - low memory support low ram: 0 - 011e0000 -> 011E0000 = 18MB RAMDISK: [mem 0x00580000-0x0062bfff] 62BFFF-580000 704511 -> ramdisk.img is 704152 compressed, where does it get unpacked to Memory: 12984K/17912K available (1990K kernel code, 446K rwdata, 532K rodata, 204K init, 148K bss, 4928K reserved, 0K cma-reserved) -> roughly 6 MB of kernel space Unpacking initramfs... -> right, where to, or does it live in the cache only? Freeing initrd memory: 688K -> is this our ramdisk? we didn't to switch_root yet - tcc is almost usable inside the i486 system, just some pathes are wrong, this small example works fine: mkdir -p /home/abaumann/projects/i486tcc-linux/build/stage1 /home/abaumann/projects/i486tcc-linux/build/stage1 ln -s /include . ln -s /lib . cat test.c < #include #include int main( int argc, char *argv[] ) { puts( "hello" ); exit( EXIT_SUCCESS ); } EOF i386-tcc -c -o test.o test.c i386-tcc -static -o test test.o => recompilation of tcc inside the 486 system might be a problem because of the build system - modern Xorg ist out of reach (and old devices DO have good framebuffer support!), Xvesa from tinyx doesn't build because of missin vm86 stuff (presumable for mode timing), this might be because we use musl? Xfbdev builds, vesafb only works when started at boot time (not as module) which is annoying. - minimize fonts - Xfbdev :0 -fp /share/X11/fonts - "Could not init font path element ../share/X11/fonts/, removing from list" https://www.x.org/releases/X11R7.5/doc/libXfont/fontlib.html => the pathes must be absolute and no errors in any file whatsover! Only BCF, PCF are supported - font handling when installing, currently in local/share/fonts/X11, we should generate them from BCF files - rxvt: can't load font "7x14" - we need to set the video mode for the floppy loading mode with rdev, vga=? seems not to be used when booting via Linux real mode boot mode (qemu_direct works fine with vga=?) - notion: keyboard XCB missing in tinyxlib? tcc: error: undefined symbol 'XkbKeycodeToKeysym' - notion: some dummy Xrandr implementation, but as long as we don't use the mod_xrandr module, we should be fine (tinyxlib has some Xrandr function stubs for a possible extention - similar like XCB - but the implementation is somehow missing) - notion: default font contains all glyphs and not only the ISO-8859-1 ones - notion: more fonts (for status bar? menues?): 2023-07-29 21:14:41 WARN /notion/../font.c:184: de_load_font: Could not load font "-*-helvetica-medium-r-normal-*-14-*-*-*-*-*-*-*", trying "fixed" 2023-07-29 21:14:41 WARN /notion/../font.c:184: de_load_font: Could not load font "-*-helvetica-medium-r-normal-*-12-*-*-*-*-*-*-*", trying "fixed" 2023-07-29 21:14:41 WARN /notion/../font.c:184: de_load_font: Could not load font "-*-helvetica-medium-r-normal-*-17-*-*-*-*-*-*-*", trying "fixed" => de/font.c if(ioncore_g.use_mb), if not use_mb, then is should use fixed => this depends on the LANG setting supposedly - notion: notion: XSupportsLocale() failed. notion: Reverting locale settings to "C". 2023-07-29 21:14:40 INFO /notion/../ioncore.c:613: ioncore_startup: Starting Notion notion: Failed to open input method. >> mod_xrandr: Unable to initialise module. >> Stack trace: 0 [C]: in 'load_module' 1 mod_xrandr.lua:5 [Skipping unnamed C functions.] 4 [C]: in 'dopath' 5 /etc/notion/cfg_defaults.lua:15 [Skipping unnamed C functions.] 8 [C]: in 'dopath' 9 /etc/notion/cfg_notion.lua:115 [Skipping unnamed C functions.] sh: xmessage: not found notion: Caught signal 2. Dying. Segmentation fault => xrandr is a dummy module (but the shared library should be loaded by lua?) => xmessage: do we really want to install that Xorg app? - wordgrinder: segfaults in luapeg on first script, probably a static linking issue or a tcc issue? - rxvt: 2.7.10 doesn't link statically and has a rxvt-2.7.10 additionally to rxvt. TODO: we use rxvt-2.6.4, why exactly? - real hardware (Unisys): softirq kernel panic Starting network via DHCP.. ne ne.0 (unnamed net_device) (uninitialized): NE*000 ethercard probe at 0x300: 00:00:e8:cd:05:88 ne ne.0 eth0: NE2000 found at 0x300, using IRQ 10. random: crng init done ------------[ cut here ]------------ NETDEV WATCHDOG: eth0 (ne): transmit queue 0 timed out 444 ms WARNING: CPU: 0 PID: 0 at net/sched/sch_generic.c:525 0xc02921bb Modules linked in: ne 8390p CPU: 0 PID: 0 Comm: swapper Not tainted 6.4.7 #1 EIP: 0xc02921bb Code: 3f c0 00 75 2a 89 45 e8 c6 05 1f 2f 3f c0 01 8b 45 ec e8 a7 e9 fd ff 8b 55 e8 52 56 50 ff 75 ec 68b EAX: 0000003d EBX: c112d234 ECX: 00000066 EDX: 00000000 ESI: 00000000 EDI: c13fd680 EBP: c0811f60 ESP: c0811f34 DS: 007b ES: 007b FS: 0000 GS: 0000 SS: 0068 EFLAGS: 00210282 CR0: 80050033 CR2: b7fe1080 CR3: 008b8000 CR4: 00000000 Call Trace: - also real hardware (Unisys): no graphic VESA, only text-VESA modes, missing => cirrus framebuffer drivers at boot time (cannot be a module)? => same behaviour with cirrus in qemu and 86box, enabling a PCI bus suddently gives us cirrusfb messages and a working /dev/fb* (not more modes though) - ${BASE} and @@BASE@@ with sed are both used to pass the build pathes to builds in patches. ${BASE} sounds like it is working more by accident.. - 'cd DIR' used everywhere without 'cd DIR || exit 1' - can we do a crunch binary for X apps, they get 400k when linked statically with X libraries1 - depth 16 -> 24 on Xfbdev? - meh segfaults with the test.qoi image from the meh branch in qemu, works on the host - slock runs but cannot unlock the screen (shadow issue?), also we need a hard-coded nobody and nogroup user and group - qemu on Arch fails to find a vgabios.bin in the local directory? => cp /usr/share/qemu/vgabios-isavga.bin vgabios.bin - some man pages are in /man, some in /share/man - non-English pages notion.1 in cs - we have a non-empty sbin - building for the first time blocks root.img in losetup, doens't create ramdisk.img, etc. races? missing trap in build.sh? - makewhatis /share/man not called on installation of root.img, results in warnings when issuing 'man xxx' - too big 6x13-ISO8859-1.pcf.gz: 4k on host, 64k in root.img. Most likely we package too many charset variants - "Linux as a Boot Loader - Peter Chubb", kernel assumes to have space after it for decompression in address space and overwrites the ramdisk, explains our fiddling with loading addresses and random crashes till they no longer overlap sessions. - suckless: - sup: not really suckless, sha256 summing and listing, daemonizing should be separate binaries in Unix sense - sandy: not really stable and maintained, it seems - static: minial cgit version with static pages tools ----- - util-linux, version 2.17: rdev, for setting the video mode in real mode load mode - bdftopcf 1.1: for converting BDF to PCF missing things -------------- We miss some things currently, must see where to get them from somwhere or whether we add them to existing minimal packages (or temporarily to abase). - abase (we already started to add some small tools) - more - hexdump - sbase - domainname (hostname exists) - ubase - fdisk - wall - ibase (this is a hypothetical thing, internet base tools, we will put them into abase for now) - ip - ping - obase: - chio (BSD I/O), stty - csh and ksh (we have oksh) - mt (tape management) - pax (archives, we have tar) - rcp/rsh (old, insecure) - systrace (as truss or strace are not really nice to build) => could be tricky to build, we don't want bmake too The comments say: "At least the ALIGN-macro is defined only meaningfully for x86_64." - 9base - ascii - awk - bc and dc - diff - mk (instead of samu) - yacc - troff - others - byacc - bison - minimal languages - lua: is already minimal and easy to build, luajit would be nice to have for lua 5.1 - microperl/miniperl https://www.foo.be/docs/tpj/issues/vol5_3/tpj0503-0003.html seems not really be working anymore after perl 5.20 microperl 5.20 segfaults - micropython: https://github.com/micropython/micropython Their definition of micro is quite different from mine in terms of size and library dependencies) I also doubt it works with tcc, needs python3 on the host to build Some todos to sort out ---------------------- - zlib - use libz or zlib-ng - st/rxvt - https://opensource.com/article/19/10/why-use-rxvt-terminal - tmux/screen - joe/vis - vesa/frambuffer other approaches ---------------- linux-486 --------- nice ideas, preinit decompresses lzma compressed files, uses the chromium Linux syscall wrapper to be a standalone binary (as the C library is compressed too). preinit/init.c preinit/linux_syscall_support.h no ramdisk, decompresses directly into / (which is the boot root without an initial change_root?), so is this in RAM? Uses buildroot for the toolchain. Boot loader is LILO. C-library is uclibc. There are 486-related kernel patches, but it runs without issues on 6.x on a real machine, so why? I really like the shell scripts to build the system, they are extremely simple. gray486linux ------------ Maybe the final system is nice, but the Nix build system is not.. 0001-WIP-patch-kernel-to-skip-endbr32-instruction-on-real.patch interesting approach to ignore endbr32 without the help of fixing gcc/binutils (we should check again, why we got it inside busybox, musl and tinycc don't to this kind of stuff). Using dropbear could be a good idea. aboriginal ---------- miniconfig.sh hush as init shell highly portable lots of patches, we must see if those are still relevant the comments in patches and comments are really funny and helpfull. also nice autoamitzation, smoke tests using qemu and report generation Firmware Linux, Rob Landley, "the 7 packages to build whole of LFS on top". The idea of making just 7 packages cross-compile and compile the rest inside the emulation. "we cross-compile for you so you don't have to". 'mkroot' is the successor. 'toybox' is the modern variant for minimal user land. oasis ----- by Michael Forney, author of cproc statically linked, the Lua build thing is weird. He also rewrote ninja as samurai in C (which is nice). Also with QBE and cproc the user land can be built with a small C compiler. QBE is currently lacking IA-32 support, so we can not just rebuild oasis for our purpose (though we could write a IA-32 QBE code generator). The idea of using git for installation feels strange to me. Also lua scripting for all this software introduces high maintaince costs. the package list was more or less an inspiration for our list of packages: musl instead of glibc sbase instead of coreutils ubase instead of util-linux oksh instead of bash sdhcp instead of dhclient or dhcpcd vis instead of vim or emacs netbsd-curses instead of ncurses mandoc instead of man-db samurai instead of ninja those packages we are still considering: pigz instead of gzip bearssl instead of openssl byacc instead of bison perp and sinit instead of sysvinit or systemd netsurf instead of chromium or firefox embutils for our abase version of: domainname install morpheus -------- statically linked Plan 9 mk build system x86_64 and i486 builds quite old (kernel 3.10.32) git://git.2f30.org/hbase (Heirloom as alternative base for some Unix tools) text-based tools like abduco and dvtm for easy text console multiplexing and window manager, no X tinyalsa sounds interesting too how is id prefering tools over other tools? stali ----- uses suckless tools everything statically linked sabotage -------- They use an old gcc without a lot of depenendencies (C++, libraries, etc.) to build the toolchain for stage 0. This is similar to us using tcc. This is the first distro based on musl which made musl much more suitable to build a Linux userspace upon. Since alpine linux switched to musl, sabotage seems to be not heavily developed anymore. Minimal Linux Live ------------------ No toolchain, so we depend on the hosts toolchain, this breaks when building too easily. Kernel 5.18 no longer compiles with newer gcc versions. links ----- - Other similar distributions or projects - Linux-i486 - https://github.com/tcsullivan/linux-486 - Aboriginal Linux/Mkroot/miniconfig: - https://github.com/landley/aboriginal - https://github.com/landley/mkroot - https://landley.net/aboriginal/documentation.html - http://landley.net/aboriginal/history.html - Morpheus - https://git.2f30.org/morpheus/ - Oasis - https://git.sr.ht/~mcf/oasis - https://github.com/oasislinux/oasis - Stali - https://sta.li - https://pub.nethence.com/xen/guest-stali - Graylinux - https://github.com/marmolak/gray386linux - https://github.com/marmolak/gray486linux - https://linuxnews.de/alte-schaetzchen-linux-optimiert-fuer-386-und-486/ - Sabotage Linux: - http://sabo.xyz/ - https://sabotage-linux.neocities.org/ - https://sabotage-linux.neocities.org/blog/11/ - https://github.com/scanlime/avc-edge-linux (disappeared) - others minimal linux: - https://www.glaucuslinux.org/ - https://github.com/ZilchOS/bootstrap-from-tcc: this is more for having a minimal tcc-based binary core for bootstrapping - https://github.com/GataOS/musl-tcc.git: especially interesting to have a look at the patches for making musl build with tcc (we sort of yoloed the complex and floating point stuff) - https://github.com/oriansj - https://bellard.org/tcc/tccboot.html - https://github.com/michaelforney/ - https://ella-0.github.io/ - https://github.com/landley/mkroot - https://github.com/richfelker/musl-cross-make.git - https://github.com/OSPreservProject/oskit - https://github.com/oriansj/M2-Planet.git - https://github.com/MiguelVis/mescc.git - https://gitlab.com/janneke/mes.git - https://www.youtube.com/watch?v=iU0Z0vBKrtQ (Michael Matz, tintycc contributor, tccboot revived, tcc hacked did it on kernel 4.6, not published) - https://git.mills.io/prologic/ulinux - https://www.plop.at/en/ploplinux/downloads/full.html - https://git.savannah.nongnu.org/git/stage0.git - https://github.com/firasuke/awesome - https://github.com/firasuke/mussel (Firas Khalil Khana) - https://github.com/pikhq/bootstrap-linux - https://www.glendix.org/: Plan9-like distro - https://github.com/ivandavidov/minimal: Minimal Linux Live - https://github.com/yeokm1/gentoo-on-486 - https://distr1.org/: Michael Stapelberg, alternative package management for Linux - https://kisslinux.org/ - https://git.sdf.org/midfavila/musl-tcc - other minimal floppy system - https://github.com/gotoco/PE_Bootloader_x86 - https://github.com/fysnet/FYSOS/ (smallc, boot loader in C for unreal mode) - https://github.com/alexfru/BootProg - http://dc0d32.blogspot.com/2010/06/real-mode-in-c-with-gcc-writing.html - https://wiki.osdev.org/Disk_access_using_the_BIOS_(INT_13h) - https://en.wikipedia.org/wiki/INT_13H#INT_13h_AH=08h:_Read_Drive_Parameters - https://stanislavs.org/helppc/dbt.html - http://vitaly_filatov.tripod.com/ng/asm/asm_024.9.html - https://web.archive.org/web/20160619063203/http://www.ctyme.com/intr/rb-0621.htm - https://students.mimuw.edu.pl/SO/Linux/Kod/arch/i386/boot/bootsect.S.html - other minimal toolchains - https://github.com/pts/pts-tcc - C Compilers - https://repo.or.cz/w/tinycc.git - https://blog.stephencleary.com/2010/10/implementing-gccs-builtin-functions.html: on __builtin_clz - C Libraries - https://www.fefe.de/dietlibc - https://www.monperrus.net/martin/compiling-tcc-with-dietlibc - https://dietlibc.fefe.narkive.com/mf2Z2Db3/tcc: sadly has issues and was never designed to use tcc in the first place - tinycc and musl - https://wiki.musl-libc.org/design-concepts.html - https://www.mail-archive.com/tinycc-devel@nongnu.org/msg08533.html - https://lists.gnu.org/archive/html/tinycc-devel/2015-01/msg00135.html - https://www.mail-archive.com/tinycc-devel@nongnu.org/msg10201.html - https://www.mail-archive.com/tinycc-devel@nongnu.org/msg10196.html - Terminal stuff - https://github.com/sabotage-linux/netbsd-curses/ - https://www.leonerd.org.uk/code/libtermkey/ - http://git.musl-libc.org/cgit/uuterm/tree/README - Editors - https://github.com/martanne/vis - https://joe-editor.sourceforge.io/ - https://ex-vi.sourceforge.net/ - https://git.sdf.org/midfavila/mse - Kernel - Tinification - https://weeraman.com/building-a-tiny-linux-kernel-8c07579ae79d - https://tiny.wiki.kernel.org/ - http://events17.linuxfoundation.org/sites/events/files/slides/tiny.pdf - https://z49x2vmq.github.io/2020/12/24/linux-tiny-qemu/ - http://elinux.org/Kernel_Size_Reduction_Work - https://stackoverflow.com/questions/27941775/what-is-the-minimum-amount-of-ram-required-to-run-linux-kernel-on-an-embedded-de - Boot process - ramdisks - https://linuxlink.timesys.com/docs/startup_overview - https://people.freedesktop.org/~narmstrong/meson_drm_doc/admin-guide/initrd.html - https://firasuke.github.io/DOTSLASHLINUX/post/booting-the-linux-kernel-without-an-initrd-initramfs/ - https://docs.kernel.org/admin-guide/initrd.html - https://landley.net/writing/rootfs-programming.html - https://github.com/chris-se/tiny-initramfs - device management - https://git.suckless.org/smdev - init systems - git://git.suckless.org/sinit - https://github.com/bbenne10/rc - https://github.com/hut/minirc - https://omaera.org/wlog/tech/runit_init - https://forums.gentoo.org/viewtopic-t-1000292-start-0.html - https://busybox.busybox.narkive.com/PBeyeZOM/custom-initrd-using - https://github.com/shiftypenguin/uinit - service/daemon management - http://r-36.net/scm/svc/log.html - minimal userland - shells - POSIX compliant - https://github.com/ibara/oksh: good compromise of good and small code base and functionality - https://yash.osdn.jp/index.html.en - dash: something bothered be there in the build process (TODO: what?) - minimalistic to the extreme: - https://github.com/hacheigriega/sshell.git: very minimal, more an education project - the 'S' shell - suckless - git://git.suckless.org/sbase - git://git.suckless.org/ubase - https://git.andreasbaumann.cc/cgit/abase/ - other minimal user lands (we don't use for now) - https://busybox.net/ - https://landley.net/toybox/ - https://tools.suckless.org/9base/ - https://github.com/leahneukirchen/obase - https://beastiebox.sourceforge.net/ - https://www.fefe.de/embutils/ - https://heirloom.sourceforge.net/ - gopher://bitreich.org - https://git.2f30.org/hbase/log.html: Heirlom - https://www.glendix.org/: we could also go into the direction of using plan 9 things (as they tend to be more minimalistic): - git://git.9front.org/plan9front/plan9front - git://git.suckless.org/9base - https://github.com/ecki/net-tools.git - https://www.reddit.com/r/suckless/comments/n14xnq/can_sbase_replace_busybox/ - https://exactcode.com/opensource/minised/ (Eroc S. Raymond, maintained by Rene Rebe) - https://stackoverflow.com/questions/4327942/non-buffering-stdin-reading - minimalism - https://wiki.installgentoo.com/wiki/Software_minimalism - https://suckless.org/philosophy - https://github.com/173duprot/harm-less - Link collections of interesting software - https://suckless.org/rocks/ - https://suckless.org/other_projects/ - https://github.com/firasuke/awesome - https://wiki.musl-libc.org/alternatives.html - http://bashismal.blogspot.com/2011/10/unbloated-resources-in-c.html - cruching binaries - https://github.com/ryao/crunch - http://www.bsdnewsletter.com/2003/09/Features102.html - TTY stuff - controlling terminals: /dev/console on serial and tty0 has no tty! - https://debianforum.de/forum/viewtopic.php?t=169622&start=15 - https://www.linuxquestions.org/questions/linux-from-scratch-13/%5Bsolved%5D-failing-to-open-dev-tty-4175685290/ - https://unix.stackexchange.com/questions/170063/start-a-process-on-a-different-tty - historical stuff - https://www.elinux.org/Busybox_replacement_project - qemu - https://en.wikibooks.org/wiki/QEMU/Devices/Network - https://virtuallyfun.com/2020/07/31/using-qemu-in-2020/ - nbd - https://patchwork.kernel.org/project/linux-block/patch_20190612163144.18486-2/roman.stratiienko@globallogic.com/ - http://patraulea.com/nbd-http/ - https://github.com/NetworkBlockDevice/nbd/ - https://patchwork.kernel.org/project/linux-block/patch/20190612163144.18486-2-roman.stratiienko@globallogic.com/ - debugging - https://www.kernel.org/doc/html/v4.15/dev-tools/kgdb.html - https://ja.nsommer.dk/articles/linux-and-tiny-c-compiler-in-the-browser-part-one.html - X - tiny X - https://github.com/idunham/tinyxserver and https://github.com/idunham/tinyxlib (puppy linux): has nice simple makefiles - https://github.com/tinycorelinux/tinyx, is newer, but requires autoconf to build - alternatives: - https://github.com/michaelforney/velox, https://github.com/michaelforney/swc: but this is wayland - http://www.microwindows.org/ - KDrive (Tiny X, TinyX): Keith Packard - https://www.pps.jussieu.fr/~jch/software/kdrive.html - https://github.com/Earnestly/sx: alternative xinit/startx - http://www.microwindows.org/ - http://forum.tinycorelinux.net/index.php?topic=16299.0: fork of tinyx for tinycorelinux - X terminal - xterm: bloat - st, requires Xft, requiring fontconfig, freetype2, etc. can eventually be redone to use old font library? - rxvt: simple and uses normal bitmap fonts (old font library) - X fonts - https://www.x.org/archive/X11R6.8.0/doc/fonts2.html - https://www.x.org/archive/X11R7.5/doc/man/man1/mkfontdir.1.html - https://www.x.org/archive//individual/font - X apps and stuff - https://www.x.org/archive/individual/app/ - terminal multiplexing - tmux: newer - screen: looks a little bit aged, might habe more features but also more code - window manager - dwm/dmenu or notion3 - notion3/notion: uses lua, loadable modules, etc. has xinerama/xrandr support, this is all a little bit too much - ion1: more suckless, maybe patchable to do what we want https://github.com/apenwarr/ion1 - debugging - automatic serial without getty - https://github.com/Angstrom-distribution/meta-linaro/blob/master/recipes-core/auto-serial-console/auto-serial-console/auto-getty - suspend/hibernate - https://docs.kernel.org/power/s2ram.html - philosophical stuff - 1983 Rob Pike made a presentation titled ‘UNIX Style, or cat -v Considered Harmful’ - http://harmful.cat-v.org/cat-v/ - hardware-specific stuff - Unisys CWD-4002 - http://www.dosdays.co.uk/topics/Manufacturers/cirrus_logic.php: graphic chip - graphic programming - https://www.svgalib.org/ - https://benpfaff.org/projects.html: BOGL graphics library for framebuffers performance ----------- Building i486tcc-linux 5 minutes userland including cross-toolchain 15 minutes kernel 1 minute ramdisk/floppy creation framebuffer ----------- insmod /lib/modules/drivers/video/fbdev/core/cfbcopyarea.ko insmod /lib/modules/drivers/video/fbdev/core/cfbfillrect.ko insmod /lib/modules/drivers/video/fbdev/core/cfbimgblt.ko insmod /lib/modules/arch/x86/video/fbdev.ko insmod /lib/modules/lib/fonts/font.ko insmod /lib/modules/drivers/video/fbdev/core/fb.ko insmod /lib/modules/drivers/video/fbdev/cirrusfb.ko insmod /lib/modules/drivers/video/vgastate.ko insmod /lib/modules/lib/fonts/font.ko insmod /lib/modules/drivers/video/fbdev/vga16fb.ko mknod /dev/fb0 c 29 0 cat /sys/class/graphics/fb0/name Unsorted links -------------- - framebuffer - https://higfxback.github.io/ - https://github.com/rothgar/awesome-tuis - https://dev1galaxy.org/viewtopic.php?id=5400 - https://directfb2.github.io/ - https://github.com/directfb2/DirectFB2 - https://github.com/8l/fbui - https://manpages.ubuntu.com/manpages/xenial/man1/fbi.1.html - https://kevinboone.me/linuxfbc.html?i=1 - https://benpfaff.org/projects.html: vga16fb - fonts - https://github.com/ntwk/vga-font Real hardware ------------- Unisys CDW-4002 --------------- cd lib/modules/drivers/net/ethernet/8390 insmod 8390p.ko insmod ne.ko io=0x300 irq=10 ifconfig eth0 up sdhcp