Realmode Assembly – Hello World Bootloader | Hacker News Reader

Realmode Assembly – Hello World Bootloader

0x00sec.org

89 pointsmiobrien11 comments

A lot of people today dismiss realmode as being old and irrelevant for writing OSes, but IMHO the constraints of the environment are not going to be relevant if you're just starting out, and especially if you're working in Asm --- even if you stay within a single segment and ignore segmentation etc., you're "limited" to "only" 64KB of RAM, but that's already quite plenty to experiement with; keep in mind that the original Unix, complete with shell, utilities, kernel, etc., ran on a PDP-11 with a 64KB total address space.

I used to teach an Asm course, and a lot of my students were wondering why we started with 16-bit, single-segment DOS COM programs --- being used to multi-KB or MB apps that don't do much, they thought it'd be impossible; and yet came away with a completely new perspective on how far away that 64KB limit seems, even if you're writing moderately complex programs.

Because that's like teaching perl cgi in a web dev course. Sure you can do a lot, but that's not relevant to today's tech, does not give them any practical experience on real world architectures, and is very limited as to what they can achieve if they want to go further. As a final note, I find it rather annoying that the web is flooded of articles on how to create a basic bootloader/kernel in real mode and using the BIOS, while there is a clear lack of good introductory articles on how to create minimalistic kernels using a stack like Core boot, uefi, TianoCore, long mode

Agree about the lack of writeups for modern architectures.

For UEFI, I believe this article here has been posted on HN before: http://www.rodsbooks.com/efi-programming/hello.html

A quick result yields various results across the internet: https://duckduckgo.com/?q=uefi+bootloader+hello+world&t=ffip...

Because that's like teaching perl cgi in a web dev course.

"perl cgi" can be not used at all, but today's x86 machines still boot in realmode and you can already do a lot of interesting things there.

while there is a clear lack of good introductory articles on how to create minimalistic kernels using a stack like Core boot, uefi, TianoCore

That could be explained by the massive increase in complexity that UEFI entails compared to BIOS. Presumably it is only after you have gotten everything working with the BIOS that will prepare you to tackle UEFI; but speaking as someone with many years of experience with the former, and even read the BIOS source listings from the original IBM manuals, the UEFI spec was still ridiculously dense and if I were writing my own OS I would definitely choose the BIOS.

Related reading from someone who has written his own, now rather famous, OS: http://yarchive.net/comp/linux/efi.html

How much more complex would it be to set up protected mode and use it in assembly? Genuinely curious.

Depends. If you use a bootloader like grub that will put you in protected mode (The GDT is a bit undefined, but in general you can probably get away with just leaving it for simple programs). However, when you go to protected mode you lose access to the BIOS interrupts (and DOS interrupts if you're using them) and gain access to other features, so the range of things that are easy vs. hard changes a fair amount.

But if you're just planning on writing very simple text programs in ASM, then the only things that are really relevant is the text display, keyboard, and possibly disk/floppy. The text display is exactly the same (assuming the GDT from grub places it at the same location), but the keyboard is a fair amount more involved, and floppy is basically impossible (without writing a fairly involved driver). An IDE driver is surprisingly simple though (But still requires an interrupt, as does the keyboard). Both of these things are basically functions calls when you still have the BIOS around, so it does create a bit of a problem.

Actually it depends on how far you want to go. You can jump to protected mode with a few lines more. But you won't have a very interesting exercise.

Basically, one would have to setup multiple registers like the GDT which sets the protection on different memory ranges (hence protected mode). Once you have your GDT setup, set a few bits in CR registers and execute a long jump to the code segment of the GDT (provided you have instructions loaded there).

Then once in protected mode, everything changes. The system allows you to set interrupts habdlndlers through the LIDT, to enable memory paging and you'll have to reload the GDT again. But no more access to the convenient BIOS functions.

You also set up the PIC of your system to get a ticking clock for scheduling your kernel operations. Once you got all that is where the fun begins and you can start implementing memory managers and allocators, writing a disk operations layer and implement (or write your own!) filesystem.

OS Dev is truly a fascinating subject because there's so much to learn. Even more with 64-bit, and multiple core initialization and usage.

I don't know how all of this works on ARM, but I might order myself an experimentation board to get my hands on.

The traditional ARM answer is that at a low level and at bootup the details are all very board specific -- what kind of BIOS equivalent you have (if any), where it starts you off, whether there's a bootloader like u-boot involved, where devices like the serial port are: all these things vary by board. For 64-bit ARM things are moving a bit more towards a standard setup with UEFI. In general, ARM assumes that the initial boot code will be customised for the specific chip, so unlike x86 which starts out looking like an 8086 until you flip it into protected mode, a 64-bit ARM CPU will start out in 64-bit and you can get at the backwards-compatibility 32-bit support by dropping down from 64-bit to 32. Back-compat is for applications, not OSes and bootloaders :-)

this is a terrible hack, but if you leave the bios areas alone, you can make a system call like machine that downshifts back into real mode, makes the call, and shifts back up into protected mode (or all the way back up into 64 bit mode)

its not _really_ worth it, but if you're just noodling around and want to send packets or write to disks you don't have to write drivers. of course you have to have dedicated buffers visible in both spaces.

the only other reason i can think of to use real mode is to actually explore the segment model rather than just treating it as trash you have to walk over to get something up from reset...but it would probably be better just to invent some kind of segmented virtual machine

About this much more complex -- https://github.com/charlesap/bootpager (example legacy boot block that jumps into protected mode, enables paging, and pages in and executes arbitrary code, no bios used.) As mentioned above the trick is what to do once you're there. And it is a UEFI world now, might as well just write 64-bit PE kernel and be done with it.

This reminds me a little of Broken Thorn's OSDev tutorial series, but that one is far more in depth.

Reply on news.ycombinator.com