According to Wikipedia, GRUB was released in 1995. By that point Linux and xBSD existed for several years. I know early Unix versions were tied to hardware in the 70s and 80s, but Linux and xBSD were free to distribute and install. Which begs the question how would you boot Linux back then? Were distributions shipping with their own implementations of bootloaders?
-
33Umm... You weren't around when LILO was the only Linux bootloader? And I've _never_ used LILO nor Grub on my BSD systems. Which one are you interested in? See e.g. [`biosboot(8)`](https://man.openbsd.org/biosboot.8). – Kusalananda Jan 07 '19 at 10:45
-
9@Kusalananda Unfortunately, I cared more about toys and drawing Ninja Turtles than pipes,execs, and shells back then :) I'm interested in general history, not specific bootloader. From the page you've linked I see OpenBSD has `biosboot` for two architectures, i386 and amd64. Does that mean OpenBSD specifically had to target the architectures instead of having one, unifying tool ? – Sergiy Kolodyazhnyy Jan 07 '19 at 10:50
-
1The first-stage bootloader would be different for each architecture (only i386 and amd64 has a "bios" BIOS anyway). Have a look at NetBSD if you're interested in more exotic architectures than the bog standard PC. – Kusalananda Jan 07 '19 at 10:58
-
1@SergiyKolodyazhnyy: there are still boot loaders for other architectures, most notably U-Boot, for embedded systems. – ninjalj Jan 07 '19 at 12:18
-
3@Kusalananda I don't think LILO ever was the only Linux bootloader. As far as I know the loader built into kernel images predates LILO, and by the time support for the built in loader was discontinued there were at least a handful of other bootloaders around. – kasperd Jan 07 '19 at 18:12
-
1@SergiyKolodyazhnyy All OSes have to target specific architectures. It just so happens that i386 and amd64 have the same underlying instruction set so some bootloaders decided to target both. Linux on the Raspberry Pi for example does not (cannot) use Grub. – slebetman Jan 08 '19 at 01:17
-
@slebetman Interesting to know about Raspberry, I'm actually interested in embedded systems. As for OS architectures, I knew that - I've been using Linux on both amd64 and i686 ( which gave me bunch of pains during installation, though I think it's more of a machine issue, than arch ). I just never realized bootloaders had to take arch into consideration as well. – Sergiy Kolodyazhnyy Jan 08 '19 at 02:13
-
3LILO was the default bootloader for a lot of distros up until the beginning of 2000s – phuclv Jan 08 '19 at 02:39
5 Answers
The first Linux distribution I used back in the 90s (Slackware 3.0 IIRC) used LILO as a bootloader. And many distros used LILO for years even when GRUB was becoming the "default" bootloader.
Moreover, in the early years of Linux it was common to boot Linux from another OS (i.e. DOS or Windows) instead of relying on a bootloader/dual booting. For example there was loadlin.
Don't forget Syslinux, which is a simpler boot loader often used for USB self-bootable installation/recovery distros. Or Isolinux (from the same project) used by many "Live" distros.
Keep in mind that today GRUB can be used to load many operating systems, while LILO was more limited, and specifically targeted at Linux (i.e. LInux LOader), with some support for dual booting to Windows.
GRUB is very useful for dual/multi booting because of its many configurable options, scripting capabilities, etc...
If you just want a single OS on your machine "any" (i.e. whichever bootloader is the default for your Linux/BSD distribution) should be enough.
- 4,127
- 2
- 29
- 30
-
I know I used to dual boot back in the lilo day and load Windows with it. I don't remember how anymore (it's been ages) so maybe it just passed me on to some Windows tool to boot the OS. – terdon Jan 07 '19 at 10:53
-
@terdon yes, I dual booted with LILO, too. But it was quite limited (i.e. no UEFI support IIRC) and sure it gave me LOTS of headaches back then (with frequent reformatting, too). GRUB is **way** better ;) – Daniele Santi Jan 07 '19 at 10:55
-
6@MrShunz: there was no UEFI back then. Booting windows was just a matter of adding an entry of e.g. `other=/dev/hda1` with `table=/dev/hda` to `lilo.conf`, and lilo would just transfer control to the boot sector at hda1, knowing the partition table would be at hda. – ninjalj Jan 07 '19 at 12:10
-
@ninjalj Yes, back then there was no (U)EFI. But LILO was still wide in use when UEFI became mainstream, and its lack of support for was one of the causes that made GRUB a better alternative (there was an ELILO project with support for UEFI IIRC). – Daniele Santi Jan 07 '19 at 12:21
-
For some reason, i had always assumed that LILO stood for Linux In, Linux Out! – Tom Anderson Jan 07 '19 at 12:32
-
LILO was quite simple and very very reliable IF you set it up correctly. No dependence on the filesystem at all. No moving parts so to speak. – rackandboneman Jan 07 '19 at 13:31
-
2You used to be able to get NTLDR to load LILO; see http://jaeger.morpheus.net/linux/ntldr.php; I discovered the same independently, back in the day. – Roger Lipscombe Jan 07 '19 at 13:41
-
2The downside of the LILO approach is it breaks if the location of disk of the files to load changes. In particular this means that LILO needed to be re-written to the boot location (either MBR or partition boot sector) after every kernel upgrade. – plugwash Jan 07 '19 at 14:26
-
Syslinux is even younger than Grub. The initial commit is from 1998-01-31 (s. https://repo.or.cz/syslinux.git/commit/00984eaa22e6aae6bbc9c83515e275011edab5ae) – Jochen Lutz Jan 08 '19 at 08:12
-
1@plugwash: GRUB has the same issue with its second stage file. The difference here is that 1) the "second stage" of LILO _was_ the kernel, so it was kernel updates, not LILO updates that broke things; and 2) GRUB updates include an automatic re-writing of the second stage location to the MBR (with the second stage then loading the Linux kernel, with full knowledge of the file system so kernel location does not matter). ;-) – DevSolar Jan 08 '19 at 13:54
-
1IIRC grub will if possible store a "stage 1.5" that understands the file system between the MBR and the first partition and will only resort to storing a reference to specific filesystem sectors if there is no room for stage 1.5 (or if it is installing to a partition boot sector rather than the MBR) – plugwash Jan 08 '19 at 13:55
-
-
2
LILO was the de-facto standard for booting Linux on PCs before Grub, from a very early stage (MCC, one of the first Linux distributions, used it). Various other bootloaders were used contemporaneously. Loadlin was quite common; it booted Linux from DOS, and was even used in some configurations with umsdos to host a Linux environment in a DOS file system... Another common configuration didn’t involve a bootloader at all: the kernel could boot itself from a floppy, and most Linux users kept a known-good pair of “boot and root” floppies, one containing the kernel, the other a basic root file system for rescue purposes.
There were a number of ways of using other operating systems’ bootloaders to boot Linux too; for example, OS/2’s boot manager, or Windows NT’s NTLDR.
Other systems had their own bootloaders:
- SILO on SPARC (Sun workstations and others);
- PALO on PA-RISC (HP workstations);
- YaBoot and Quik on PowerPC;
- aBoot and MILO on Alpha...
Even nowadays Grub isn’t the only bootloader you’ll see. While having the kernel boot directly from floppy is no longer very useful (I haven’t checked whether it’s still possible, assuming you can build a kernel small enough to fit on a floppy), it can boot directly from EFI (which is effectively its own small operating system designed to load other operating systems, as is Grub). On many smaller systems (embedded systems, single-board computers...) you’ll find U-Boot. (And there’s also an EFI layer for U-Boot.)
- 411,918
- 54
- 1,065
- 1,164
-
The PowerPC architecture is also interesting because some motherboards had a Turing-complete BIOS - Openfirmware (basically the Forth programming language with some preinstalled functions). This allowed booting directly from BIOS without bootloader if you know how to configure your BIOS – slebetman Jan 08 '19 at 01:21
-
Hey, just curious, NTLDR can load linux kernel directly? I heard that NTLDR can chainloader grub4dos and then load the linux kernel. – 炸鱼薯条德里克 Jan 08 '19 at 04:09
-
@slebetman: More precisely, OpenFirmware was developed by Sun for SPARC and then adopted by the PowerPC alliance (IBM, Apple, Motorola) for the PowerPC Reference Architecture, and specifically by Apple for PowerPC-based Macintoshs. One of the powerful aspects was that simple drivers could be stored inside ROM chips on the expansion cards, or in some designated boot area of a HDD, and since they were written in bytecode against a known specified ABI, they would work regardless of which CPU architecture and OS you were trying to boot. – Jörg W Mittag Jan 08 '19 at 14:07
-
E.g. you could have RAID adapter which has its OpenFirmware driver inside a ROM chip, then the OpenFirmware environment could use that driver to access the RAID, inside the RAID, there could be another driver for the partition table format, which would allow the OFW environment to find the partitions, at the beginning of each partition would be an OFW driver for the filesystem, which would allow the OFW system to find the kernel, and the kernel would have a small bootloader written in OFW bytecode at the beginning. – Jörg W Mittag Jan 08 '19 at 14:10
-
GRUB can operate in a similar way, but the difference is that all those drivers have to be written specifically for GRUB, whereas the beauty of OFW was that the device would bring its drivers with it, which means that even devices that didn't yet exist when the OFW environment was written would just "magically" work. UEFI can also operate in a similar way, but its "portable bytecode format" is essentially a subset of DOS, which is the main reason why Itaniums still need an x86 emulator. – Jörg W Mittag Jan 08 '19 at 14:13
-
@炸鱼薯条德里克 NTLDR can chainload any boot sector; so you can use it with Grub, or LILO, to boot Linux — and ISTR using a simpler boot sector back in the day, without Grub or LILO (that was nearly twenty years ago, I don’t remember the exact details; I have the setup on tape somewhere but I’d have to dig it out). – Stephen Kitt Jan 08 '19 at 14:21
-
@JörgWMittag this neat feature allowed a few Mac-flavoured PCI devices to work in Sun workstations if you didn't want to pay for an equivalent Sun-branded device. IIRC Adaptec and Matrox made such things although the Mac-flavoured version was generally scarcer than its PC sibling – bodgit Jan 09 '19 at 11:37
Up through mid 2.6 kernels, the x86 kernel was directly bootable if copied onto a floppy disk (as though it were a disk image).
This was, in fact, the original way of booting Linux.
If you look at the header of an x86 kernel today you see an error message that says booting from floppies like that doesn't work anymore.
- 1,733
- 12
- 19
-
2On the other hand, the x86 kernel is now directly bootable if given to an UEFI firmware. So there's still a stub bootloader tacked in front of the kernel, just a different type... – u1686_grawity Jan 07 '19 at 20:58
-
-
While it's primarily used for x64, the built-in stub loader appears to be compatible with 32-bit EFI if the kernel itself is 32-bit, according to documentation in the arch/x86/Kconfig. (Additionally, "On the x86 and ARM platforms, a kernel zImage/bzImage can masquerade as a PE/COFF image, thereby convincing EFI firmware loaders to load it as an EFI executable" according to the doc file.) – u1686_grawity Jan 07 '19 at 21:02
-
@grawity I looked at a recent kernel image and I saw that it starts with the two characters MZ, which signals an EXE format file. At position 510 I saw two bytes with hexadecimal values 55 and AA, which signals that this is a PC boot record. If used as a boot record it will execute from the beginning of the block, so the two bytes MZ would actually be executed as 8086 code. They probably don't do anything harmful, so it's still possible to make it work as a boot record. I don't know if there are more magic numbers in there, which I didn't notice. – kasperd Jan 08 '19 at 13:51
-
@kasperd: There probably aren't. The "EXE format file" bit has a specific purpose – that's the EFI stub bootloader (so the kernel can be an .efi file, which are just EXEs). – u1686_grawity Jan 08 '19 at 14:07
-
-
1@Joshua: I'm not sure what you mean by that. EFI doesn't actually execute this part as code. – u1686_grawity Jan 08 '19 at 14:12
-
2@Joshua what? It’s “DEC BP”, “POP DX” in 16-bit mode (EBP/EDX in 32-bit mode). But it shouldn’t be executed anyway; EFI binaries are PE files (which of course doesn’t matter if it’s written to a boot sector...). – Stephen Kitt Jan 08 '19 at 14:14
-
@StephenKitt: If you wrote the kernel to a floppy and booted it, MZ would execute and underflow the stack leaving SP pointing into BIOS data block and you've got more headers to run after that. – Joshua Jan 08 '19 at 14:23
-
1@Joshua OK, but that’s not undefined x86 behaviour in my mind ;-). (I think of “undefined x86 behaviour” as opcodes whose behaviour isn’t defined, not undefined platform behaviour.) – Stephen Kitt Jan 08 '19 at 14:26
-
@kasperd it might be you have to make sure not to enable CONFIG_EFI_STUB, when you want to boot the x86 floppy stub loader that tells you the x86 floppy stub loader does not work anymore :-). – sourcejedi Jan 08 '19 at 23:58
-
@sourcejedi: Hilariously I have a patch that puts the loader back without using the spare sectors and can handle a kernel up to 1.68mb (else out of disk). Nobody cares. – Joshua Jan 09 '19 at 00:00
-
@Joshua How would underflowing the stack end up in the BIOS data block? Isn't the stack pointer initialized to point somewhere in the top of the first 32KB (which is the minimum amount of memory a PC could have). Underflowing from there would not get anywhere near any BIOS data. – kasperd Jan 09 '19 at 00:16
-
@Joshua I am sure the 90s would love your patch to allow booting larger than 1MB kernels directly from floppy. I am afraid the 10s don't care for booting from floppy anymore. – kasperd Jan 09 '19 at 00:18
-
I started with Linux in the late 90s and as mentioned lilo was the default. If you wanted to dual boot with a DOS system, you could do a bare boot without loading stuff into HIMEM or loading CD drivers, etc. and use loadlin. For Win95 dual booting, you could make the drive bootable first with DOS, then install '95, and '95's boot loader would let you boot the DOS kernel still, and then you could use loadlin.
For dual booting with NT4, the trick was to write LILO to the / partition, then strip off the first 512 bytes using dd (dd if=/dev/sda2 of=/path/to/file bs=512 count=1) and put the resulting file where ntldr could see it and you could use it from WinNT's boot loader. The issue with doing that is when you upgraded your kernel you had to remember to repeat all the steps before rebooting, otherwise you'd have issues getting back into the Linux system. Same process worked with Win2k.
With LILO, any time the kernel was updated, you had to remember to update LILO.
With loadlin any time the kernel updated, you had to remember to copy the kernel out to the DOS partition.
One other option that is hinted at in other answers was to write the kernel directly to a floppy using dd if=/path/to/vmlinuz of=/dev/fd0 BUT the root device had to be set properly in the kernel, either at compile time or by using the rdev utility.
When GRUB came around, there was much rejoicing because you no longer had to remember to update LILO, or update LILO and re-strip off the boot info, etc. No more getting left out of your Linux system because you forgot to update the boot loader info...
- 4,870
- 1
- 9
- 20
-
Sounds like it was whole lot of work and high chance to be left with non-booting machine back then, but definitely an educational experience – Sergiy Kolodyazhnyy Jan 09 '19 at 03:02
-
@SergiyKolodyazhnyy yeah, and there wasn't such a wealth of info on the internet, or the great search engines to find it if it was there. There were several single floppy disk rescue distros that had just enough Linux to boot and fix LILO, etc. We've come a long way! – ivanivan Jan 09 '19 at 03:05
-
Running `make install` would run `/sbin/lilo`, so you didn't really have to update anything by hand (and that may still be the case, if you have `lilo` installed). That may be a matter of opinion, but I don't remember much rejoicing at `grub`, at the contrary. And `lilo` (at least its 1999 version) could dual boot windows just fine, no need of `loadlin`. – Jan 09 '19 at 19:51
And before LILO and GRUB, you had to launch it from the command line with some sort of custom bootloader utility.
As an example, the Amiga had Linux available. You had to use a command line utility called amiboot to load the kernel ELF into memory and jump to it.
Here is a video of someone using amiboot from the command line to launch linux on an Amiga 600. His StartInstall script is calling the amiboot executable. You can watch amiboot configure memory, figure out the desired load address, and pass parameters to the kernel at around 0:55.
- 111
- 1