[SOLVED] UEFI system booting from MBR partition table and GRUB legacy

Question

I'm trying to understand once and for all the process by which Arch can be booted from a system with UEFI firmware and an MBR partition table. Some of the information on the wiki seems conflictual / non-nonsensical at times. Apologies in advance if this has been answered time and time again, but I did search around and all I found was fixes to get Arch to boot rather than comprehensive explanations of the boot process.
Now, the way I would imagine it works is that it's just completely identical to the way it would work with a BIOS firmware. The UEFI firmware detects an MBR partitioning scheme (or is configured to know it's an MBR partitioning scheme), activates some "legacy" mode and executes the MBR boot code, just like a BIOS firmware would.
The wiki however, says different. From the Macbook article: "Do not install GRUB onto /dev/sda !!! Doing so is likely to lead to an unstable post-environment."?
So what is there in the MBR boot sector? Nothing?
How does the firmware know what to boot if there's no 0xEF BIOS boot partition and no Grub stage 1 in the MBR boot sector?
Also, how does installing Grub stage 1 to a partition work? Does it have to be at the beginning of the partition? Wouldn't that overwrite some existing data?
I'm especially puzzled since many guides to installing Vista on a macbook recommend simply formatting as MBR, and installing as normal, which I suppose entails having the Windows installation process write its boot code to the MBR, ie the equivalent of installing grub stage 1 to /dev/sda rather than to the /boot partition, as the Macbook article suggests.
Any input is appreciated.
P.S. I realize it's probably simpler, if I just want to dual boot Windows and Arch, to install Windows 7 in UEFI-GPT mode, let it create the EFI System Partition, and then install GRUB 2 to that partition, but I'm still curious about the UEFI-MBR boot process.
Last edited by padavoine (2012-06-06 09:35:10)

Answer 1

padavoine wrote:
CSM in UEFI firmwares do the exact same job as normal BIOS firmware.
So it's something specific to the Mac that it's able to boot from a partition's VBR while ignoring the MBR?
The reason that warning is given is because grub-legacy modifies more than just the MBR boot code region.  It can overwrite some parts of GPT header.
Not true, the instruction is given in the context of an MBR format, not in the context of a GPT format, so there's nothing to overwrite and Stage 1.5 should be safely embeddable in the post-MBR gap.
In BIOS boot (normal case in non-UEFI firmwares or CSM in UEFI firmwares) does not read the partitition table (atleast it is supposed to be dumb in this regard), it simply launches whatever boot code exists in the 1st 440-byte of the MBR region.
So again, you're saying it's specific to the Mac UEFI that it lets you choose a partition whose VBR to load, regardless of what's in the MBR?
I haven't used Macs so I can't comment on Mac firmware behaviour. But normal BIOS firmwares (legacy and CSM) launch only the MBR boot code and not the partition boot code. We need some chainload capable boot manager in the MBR to launch the partition VBR.
grub-legacy does not know anything about GPT. So when you install grub-legacy to /dev/sda, it install the MBR boot code (stage1) and stage 1.5 code to the (supposed) post MBR gap. Since there is no actual post MBR gap in GPT (which has been taken over by the header and partition table), grub-legacy does not check for GPT and it assumes the post MBR gap actually exists which is invalid in case of GPT. grub-legacy embeds the stage 1.5 code in GPT header and table region (which grub assumes to be unused post MBR gap) and thus corrupts it.
0xEF is the MBR type code for UEFISYS partition. grub stage 1 (used in grub-legacy, not in grub2) is the 440-byte boot code stored in MBR for use in BIOS boot.
That's precisely my point: with neither proper executable code in the MBR (since grub was installed to a partition, not to the MBR) nor a UEFI system partition, what does the firmware default to, and how does it know what partition to boot from?
In that case it might fallback to UEFI Shell (if it exists)  or give an error similar to the case where BIOS does not find any bootable code in 440-byte MBR region.
So even with bootcamp/CSM, the disk also needs to be MBR partitioned. So Macs use something called "Hybrid GPT/MBR" ( http://rodsbooks.com/gdisk/hybrid.html ) where the MBR table is synced to match the first 3 partitions in the GPT table.
I know what Bootcamp does, and that's not what I was referring to. I was referring to standalone Vista installs. I wasn't puzzled at the fact that they were using MBR, I was puzzled at the fact that contrary to the recommendations for the standalone Arch install on the wiki (with MBR partitioning, not GPT), they didn't do anything to try and prevent Windows from writing to the MBR.
You can't prevent Windows from overwriting the MBR region. You have to re-install the bootloader (grub2/syslinux etc.) after installing Windows. That is the reason why it is recommended to install Windows first and linux later.
Thats not true. I actually find it is much easier to install Windows UEFI-GPT using USB rather than a DVD.
I haven't done it since the only UEFI system I own has no DVD drive, but I was under the impression that it was simply a matter of choosing DVD UEFI boot in the firmware's boot menu.
format the USB as FAT32 and extract the iso to it. That it.
No, thats not it, precisely, it doesn't work out of the box with a standard Windows install USB, you need to fiddle around:
2.3 Extract bootmgfw.efi from [WINDOWS_x86_64_ISO]/sources/install.wim => [INSTALL.WIM]/1/Windows/Boot/EFI/bootmgfw.efi (using 7-zip aka p7zip for both the files), or copy it from C:\Windows\Boot\EFI\bootmgfw.efi from a working Windows x86_64 installation.
2.4 Copy the extracted bootmgfw.efi file to [MOUNTPOINT]/efi/microsoft/boot/bootmgfw.efi .
Most of the Windows isos already have /EFI/BOOT/BOOTX64.EFI file, so no need to extract the bootmgfw.efi file.
There is no difference between in BIOS booting in UEFI firmwares and BIOS booting with legacy firmware.
There has to be a difference, at least in the Mac firmware (sorry, I keep switching), since legacy firmware, AFAIK, cannot chainload a bootloader in a partition's VBR without there being some sort of "stage1" code in the MBR.
No idea about Mac EFI. Apple made a spagetti out of UEFI Spec. To actually understand how Mac firmwares work, read the blog posts by Matthew Garrett of Redhat, about his efforts in getting Fedora to boot in Macs.