-------- Original Message -------- Subject: embed linux in UEFI? Date: Wed, 9 Oct 2013 11:48:35 -0700 From: Mike Hodgin email@example.com To: firstname.lastname@example.org
I should first say that I'm not much of a programmer or expert in any special way, and that I'm aware that there are probably perfectly obvious technical reasons which elude me that prohibit and/or obviate some or all of my notions. That said, if any one of you could take the time to describe what it is I don't understand in way that I might I would be very grateful.
Personally, I've always thought that the coreboot guys had it right and that Linux in one form or another is really the ideal replacement for BIOS. As I understand it, pretty much since the advent of plug'n'play the OS's have been doing the majority of its work anyway. In a video I watched once coreboot's Ron mentioned that Linux did require BIOS for hardware intialization routines, especially where embedded controllers and ACPI is concerned, but that once power was confirmed Linux (and most others) just took the reins. Probably that's an overly-simplistic description, but that's what I took away from it.
Unfortunately due to undocumented proprietary board and BIOS designs coreboot is apparently very difficult to implement. This was especially true when BIOS was BIOS and all its routines were stored in a flash module averaging 512kb or so. The work involved in reverse-engineering an undocumented system written in assembly often specifically designed to protect trade secrets and which necessarily already incorporated any and all possible hacks and compression methods possible just to make it fit in the first place boggles my mind, especially when a single failure can so often be your only chance. I'm astounded they ever made any headway at all, much less acheived compatibility with the more than 100 boards listed on their site.
But most of us are nowhere near as capable, nor so motivated, and 100 or more compatible boards or none at all makes little difference if our hardware is not included among them, as is the case for most. But lately I've noticed some interesting goings-on in the UEFI world that indicate a sort of compromise might be possible and so I thought I'd run it by you guys just to find out how off-base I really am.
Some of the hackintosh guys have apparently managed to insert their own .DXE driver modules in several system ROMs and for at least one board have installed their own tailored bootloader enabling them to boot OSX on their own PC hardware masked with custom DSDTs directly from UEFI. Some others regularly update their own firmware option ROMs in system flash beyond any board manufacturer support. I found one script that scans an image file for possibly upgradeable modules, installs them as necessary, recompresses and/or recompiles as appropriate, and leaves the user with a flashable image file - all automatically.
It seems to me that most of these happenings can be attributed to a few core factors:
Firstly, where UEFI is concerned there are three main players: AMI's Aptio, Pheonix's I-forget-the-name, and Intel's Tiano-based images. Of course this was the case before as well, but then board vendors were a lot more likely to modify their ROMs in any number of ways that were little-understood. I'm a bit unclear on this, but I strongly suspect that both AMI and Pheonix lean heavily on Intel's Tiano codebase because of their "opensource" EDK2, even if the irony that it requires a base system such as coreboot just to operate is a little ridiculous. Anyway, with fewer sources the system has become more approachably understandable from a broader standpoint than did BIOS's more specialized days.
Adding to that approachability is the move from assembly to a c codebase. I vaguely grasp that it's actually some kind of c specialization or perverse at least in some way, but I can't put my finger on exactly what that means. In any case, it's not machine language. C's adoption alone inflated ROM images many-fold, not to mention whatever other so-called value-added UEFI services might require. This has necessarily lead to the average consumer UEFI board incorporating system flash chips of 4mbs and more just to contain it. I also attribute the change from real- to protected-mode boots to the c migration, though I'm very fuzzy on technicalities. Whereas before BIOS hacking could offer rewards like a .5mb NOR or 64kb of addressable memory, UEFI seems to bring much more to the table in that respect.
Owing perhaps a great deal to larger flash chips, board vendors seem to be following the UEFI spec in large part and are conveniently rolling their applicable drivers and configuration images into UEFI modules. Many of them do so with common toolkits provided them by the likes of AMI and Intel. Whether or not any or all of them is very good at it is beside the point; what most interests me about this is that this actually very frequently is described with words such as "common," "toolkit," and "module."
I say all of this knowing that obviously anyone on the other end of this email is likely very well aware of these things and much more besides; I only wish to summarize my understanding in case I've got something wrong and someone can help me to understand it better. As I was reading through the various forum threads and about hackintosh UEFI mods and similar that I mentioned before I came to suspect that customizing my own UEFI firmware might be within my reach. The hackintosh mods specifically seem to have mostly been made possible by some curious guy with no special understanding plugging away with a few (very possibly illicitly acquired) vendor drag-and-drop tools in a trial-and-error sort of way until he achieved the results he wanted, all the while allowing his Gigabyte UEFI DualBIOS chips to restore itself themselves as necessary when he fucked it up. The self-updating UEFI script is apparently hosted and maintained by the kind of person you'd expect on a forum; when I asked him about the hows and whys he was unclear and knew only that it worked to upgrade his Intel RAID and did so for everyone else as far as he knew and that was enough.
I read through all of this and I thought, "Hey, I'm that guy. Two of my boards have buggy UEFI ROMs that haven't seen an update in a year; I have a DualBIOS backup; ; I can illicitly acquire software if need be; I can fuck things up, too!" As I said before, I'm not a programmer; I just tinker. I follow step-by-step instructions to root my android phones and pride myself on 20-lines of shell-script that I'll sweat over for an hour until finally my zsh prompt reliably presents me with a frowny-face following a non-zero exit code. And I think I can roll linux in my system rom.
Obviously the system initialization stuff is well beyond my reach, but I expect some success at least in an attempt at adding a small, kexec-capable, EFI-stub kernel to flash. Some inspection reveals that my Gigabyte board in particular has 700kb to spare even now, and that's before I replace the full 1mb shell module it mysteriously incorporates without providing any discernible means to load. Likely there are other driver modules as well that I have either no need for at boot (such as the option ROM for the flaky consumer-grade RAID controller my board advertises), or even that the vendor shipped bundled because it could in the available storage but that are only applicable to options offered on other similar boards. I know there can be an issue with some UEFI implementations when available storage trickles close to nothing, but I should be able plan for the free 32kb or 64kb necessary to avoid that. And of course there's always DualBIOS.
So finally I'm down to the questions (if you've read this far I applaud you for your perseverence in the face of boredom and thank you for your super-human tolerance). Mostly I'd like to know what I've got wrong so far? I feel like I shouldn't be alone in considering this possibility but no amount of googling has surfaced reports of similar experiments. Though iPXE forums occasionally discuss installing a UEFI version of their loader they seem to indicate it should be done to the option ROM of a PCI card as per usual. This mailing list comes close with the recent lengthy back-and-forths about kexec and UEFI possibly mapping over the kernel after exit-boot-services is called, but neither that thread or any other I've discovered mentions in-ROM installation. I assume I must have overlooked something very obvious for it not to have seriously crossed the minds of others.
Also, if it were possible, what should I consider? (Other than the real and present danger of bricking my board, of course.) Would I need to incorporate a block and/or filesystem driver for the flash chip or would UEFI's call of the kernel in flash fully map it to memory and obviate that requirement? I've noticed UEFI has this nasty habit of byte-coding anything and everything I do in the EFI shell; a "dh > fs0:\handledump" for instance is perfectly human-readable in its shell but after boot requires a decompile. Is this signifcant?
I imagine a small kernel incorporating only the barest of necessities for IP boot and/or to find and mount a disk loaded as a UEFI driver that could then either pick up capabilities as made available by an initramfs image offered on its new mount or kexec completely into another kernel. I think it can work, do you? Have you advice? Admonishment? Ridicule? I welcome all input.
-Mike -- To unsubscribe from this list: send the line "unsubscribe linux-efi" in the body of a message to email@example.com More majordomo info at http://vger.kernel.org/majordomo-info.html
----- End forwarded message -----