[LukeShu]

Very cool hack.

But, as okanat says, if the goal is to run legacy operating systems, then modern hardware is going to be challenging in other ways too. IME the biggest thing I want PC BIOS back for is the ability to use non-GPT disk layouts, which this doesn't accomplish as, as to use it you put it on the EFI partition on your GPT disk.

[mintsuki]

By spec, UEFI is required to support both MBR and GPT, so there is literally nothing stopping you from using MBR.

[skissane]

MBR and GPT can coexist on the same disk - MBR goes in sector 0, GPT starts in sector 1. Normally the MBR in sector 0 is “protective” (it just marks the whole disk as in use), but in theory you can create a “hybrid” disk in which the MBR and GPT both describe the same partitions. I suppose you could even reserve most of the disk as one big GPT partition and then split that up into multiple MBR partitions. You’d have to be very careful editing the partition table on such a disk because standard tools might break it.

[rjst01]

In practice, whether or not this actually works can be very hit-or-miss. We've found several UEFI implementations will not consider a disk bootable if the pMBR doesn't exactly match the spec, which specifies that the 'protective' partition shouldn't be marked as bootable in the MBR partition table.

Meanwhile, other implementations will not consider the disk bootable in BIOS mode if the partition in the pMBR is not marked bootable.

[mintsuki]

Indeed! Which is why the only portable solution is to not do this.

For stuff that needs to be bootable by both BIOS and UEFI the only portable solution is to use MBR, not GPT. That means all legacy BIOS systems will boot it, and so will all UEFI systems since UEFI must support MBR.

For ISOs that need to additionally be booted off of optical media (aka ISOHYBRIDs) the story gets more complicated, but ultimately what you need to take away from that is the same: avoid GPT at all cost.

[skissane]

I wonder… does the UEFI spec mandate any particular behaviour in these cases? Does the UEFI SCT test for it?

Seems like an inconsistency which could be addressed by adding it to the spec and/or test suite.

I guess the other thing I don’t know, is whether there is any actual real world pressure on firmware vendors to pass the test suite.

[LukeShu]

Too late to edit: non-GPT, non-MBR

For instance, whole-disk btrfs. Or old BSD partition schemes.

[mintsuki]

That is a very niche use case, and also won't really work on most real BIOS machines as they actually check for a proper MBR and/or BPB (and no GPT), which is also why GPT on BIOS support is very spotty, since a lot of BIOSes will detect the GPT and refuse to boot, or refuse to boot in legacy (CSM) mode.

In any case that is far beyond the original scope of booting old PC OSes, which MBR support alone serves really well (99.9% of the way there, really), which is why I assumed by default you were thinking of MBR, not some other weird scheme.

[superkuh]

GPT is not UEFI. It's just the format of the storage device. I have many computers with GPT formatted disks with actual master boot records at the start of the disk that boot MBR on both original BIOS machines and CSM BIOS machines.

MBR is two things with the same name. It is both the format of the storage device and describes a booting mode using a master boot record and up to 4 partitions.

GPT is one thing. It is a format for a storage device. It is the alternative to the MBR format for storage devices. It has nothing to do with (U)EFI.

BIOS or BIOS/csm are not types of formats for storage devices. They are types of boot processes. (U)EFI is another type of boot process.

You can easily mix and match boot types (BIOS/csm vs UEFI) and storage device format types (GPT vs MBR). As others have said, there may be some slight incompatibilities on some rare hardware/software configs, but mostly it just works.

[mintsuki]

I know every word of what I said in the message you're replying to. While what you said is mostly accurate, as in MBR != BIOS and GPT != UEFI, never in my original message have I stated that that is the case. And the rest of what you said is incorrect, as per my original message.

I invite you to read what I said again.

[superkuh]

Oof. On re-read I see now this sub-thread's topic was "non-GPT, non-MBR". So my "response" was ... off topic and useless. Sorry.

I am not sure what I said that was incorrect though. As I'm always looking to learn, which bit? I am currently assuming the 'rest' about GPT on true BIOS machines. I suppose I have never booted off GPT formatted disks in true BIOS machines. I always boot off MBR formatted disks and just used the GPT as storage/etc.

But I'm unclear on how this would break things since you can still have the MBR in the first sector on a GPT formatted disk. And how could the true BIOS motherboards detect GPT format on the disks when they don't know about it?

[mintsuki]

Because they do know about them. At least some of them do, of course.

Ideally, BIOS/CSM systems should not care about what type of partition table format a disk uses, and just boot by loading the boot sector (the 1st sector), and passing control to it.

In the real world, some machines will check for whether an MBR exists (by MBR meaning the partition table contained in the first sector, also called "msdos" partition table, naming varies depending on who you ask), some of them will check whether there is any partition set as active in said MBR, some of them will, indeed, check if there is a GPT (past any protective MBR), and ignore such disks - possibly because firmware writers thought that no GPT disk should be bootable on BIOS systems (Windows doesn't support it, so...), to avoid passing control to an empty protective MBR and possibly crash or hang as a result.

[Kwpolska]

How much benefit would there be in whole-disk btrfs? The space taken up by GPT and the FAT32 EFI partition is a rounding error compared to the sizes of modern disks.

[adrian_b]

I have used only whole-disk XFS SSDs and HDDs for more than a decade.

While the space wasted by GPT and partitions and by the arbitrary alignment rules used by various formatting tools is not great, I do not see any reason why it should exist. The space wasted now with GPT and UEFI is several orders of magnitude greater than it was with traditional partition tables, so eliminating it has become more attractive.

Such SSDs and HDDs are not bootable, but for me this is a desirable feature, not a bug. I boot my computers either from Ethernet (e.g. most of my servers) or from a removable USB memory.

The SSDs and HDDs with a whole-disk file system are also wholly encrypted. As in a proper encryption implementation, there is complete separation between the encrypted data and the encryption key. The encrypted disks or their hosting computers do not contain any information about decryption, unlike in many systems of disk encryption. The encrypted decryption keys can be found only on external or remote boot media, which are not normally associated with the hosting computers.

[cesarb]

> I have used only whole-disk XFS SSDs and HDDs for more than a decade.

I wouldn't recommend that. It makes the disk appear to be empty to many tools (and possibly some operating systems), which could lead to data loss. That's the reason GPT has the protective MBR: it makes the disk appear to be full to legacy tools which don't understand the GPT format.

[adrian_b]

You are right, and such disks will appear as empty to any tool or OS, but I will never insert any of my SSDs or HDDs that are used in this way in a computer that I do not control.

Windows frequently overwrites any bootable Linux drive, so it might have even less restraints with an apparently empty drive that does not have GPT structure.

Because of the risk of Windows messing with them, I also do not insert any of my bootable USB memories in a live Windows computer, even when they have GPT structure. For data interchange with Windows, I use only clean and non-bootable exFAT formatted USB memories.

[lproven]

> The space taken up by GPT and the FAT32 EFI partition is a rounding error

I am not the OP and I do not do this. However...

Yes, the disk space used is a rounding error and one far right of the decimal point.

But there is extra space used in the human brain that sets this up.

Instead of `/dev/sda1` it's `/dev/sda`. Instead of `/dev/nvme0l0p1` or whatever (I own almost no NVMe kit) it's `/dev/nvme0l0`. And so on.

This could be helpful; I can certainly imagine it simplifying things somewhat.

[nicman23]

you probably can with uefi modules