FlyGoat
VGA BIOS Option ROM
Forgive the perhaps silly question, but does this execute a VGA card's video BIOS option ROM, if present?
That is the best way (and only real way as far as I know) to obtain legacy-compatible video modes on a UEFI system (and yes, it will need a PCI video card shipping a legacy compatible video BIOS, but that's been the case for several generations of Intel/AMD systems now, even those supporting CSM natively).
Without this, the emulation of legacy video modes will never be capable enough to properly run a vast majority of legacy OSes and applications expecting proper legacy video modes/VGA stuff to work...
Thanks in advance for any kind reply.
No at the moment, but I'm trying to implement this functionality.
For now it uses VBIOS implemented by SeaBIOS unconditionally, which will draw onto framebuffer setup by UEFI GOP. This is pretty capable and VBE 3.0 compliment.
For native VBIOS, I met some issues on my first attempt, will try to work out a solution.
Some progress on this topic, I managed to get native VBIOS working with R5 230 (which only provides a legacy OpROM).
However, on RX550 it seems like PciIo->RomImage only contains EFI GOP OpROM. I'm sure the card provide a legacy OpROM as well because I can dump it in Linux driver. I think it's a policy of UEFI PCIBus driver's OpROM copy mechanism.
Reading the on card ROM is not a trivial task, as ROM BAR of the card was left disabled by the firmware, we need to allocate a range for it, which seems out of scope of EFI BootService. We may need to write a resource allocator on our own. In worst case, we may need to enlarge the low 4G PCI Hole to accommodate it.
Anyway, I'll try to make some progress.
Very happy to hear that progress is being made! I tried to look into it myself as well yesterday, but made little progress.
Even if in the end this only works for older cards, it's still better than nothing. This issue is really fundamental as far as this project goes since it's the n.1 issue that sits between a good CSM-on-EFI and an unacceptably limited one.
As I mentioned yesterday, I am more than happy to run tests for you if you want me to!
I have an idea - how about instead of SeaVGABIOS or graphics card VBIOS use fake int10 like OVMF or UefiSeven does - inserts fake int10 into memory at address C0000 - with this WinXP work OK on pure UEFI. UEFI use GOP for boot PC but WinXP use fake int10 from C0000 memory and works on standard vga.sys driver:
QEMU:
000C0000 56 45 53 41 00 03 22 00 00 C0 01 00 00 00 27 00 VESA..".......'.
000C0010 00 C0 30 00 00 00 2B 00 00 C0 30 00 00 C0 3D 00 ..0...+...0...=.
000C0020 00 C0 51 45 4D 55 00 F1 00 FF FF 4F 56 4D 46 00 ..QEMU.....OVMF.
000C0030 51 45 4D 55 20 51 58 4C 20 56 47 41 00 4F 56 4D QEMU QXL VGA.OVM
000C0040 46 20 49 6E 74 31 30 68 20 28 66 61 6B 65 29 00 F Int10h (fake).
000C0050 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
000C0060 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
UefiSeven:
000C0000 56 45 53 41 00 03 22 00 00 C0 01 00 00 00 2C 00 VESA..".......,.
000C0010 00 C0 30 00 00 00 30 00 00 C0 3A 00 00 C0 47 00 ..0...0...:...G.
000C0020 00 C0 55 65 66 69 53 65 76 65 6E 00 F1 00 FF FF ..UefiSeven.....
000C0030 55 65 66 69 53 65 76 65 6E 00 45 6D 75 6C 61 74 UefiSeven.Emulat
000C0040 65 64 20 56 47 41 00 4F 56 4D 46 20 49 6E 74 31 ed VGA.OVMF Int1
000C0050 30 68 20 28 66 61 6B 65 29 00 90 90 90 90 90 90 0h (fake).......
000C0060 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 ................
A long time ago I was doing tests with WinXP and inserted OptionROM VBIOS IBM-VGA.ROM, IBM-EGA.ROM into the bios and XP works without graphics card:
https://www.elektroda.pl/rtvforum/viewtopic.php?p=11836500#11836500
The screenshot was taken over the network using a connection to the Remote Administrator server installed on this WinXP:
Because IMO that isn't a real solution. Unless CSMWrap is meant solely to boot Windows NT, and it isn't, this is just a workaround and won't work with any real MS-DOS/9x/other stuff, as those all expect proper legacy video modes like especially text mode and friends, which are not emulatable the same way one can with linear framebuffers.
I have an idea - how about instead of SeaVGABIOS or graphics card VBIOS use fake int10 like OVMF or UefiSeven does - inserts fake int10 into memory at address C0000 - with this WinXP work OK on pure UEFI with GOP on vga.sys driver:
I think SeaVGABIOS is much more capable then UefiSeven's fake int10h solution as it do have proper text mode implementation and so on. The only thing missing is real modesetting, which is only possible with hardware specific drivers like real Op ROM or GOP.
QEMU:
000C0000 56 45 53 41 00 03 22 00 00 C0 01 00 00 00 27 00 VESA..".......'. 000C0010 00 C0 30 00 00 00 2B 00 00 C0 30 00 00 C0 3D 00 ..0...+...0...=. 000C0020 00 C0 51 45 4D 55 00 F1 00 FF FF 4F 56 4D 46 00 ..QEMU.....OVMF. 000C0030 51 45 4D 55 20 51 58 4C 20 56 47 41 00 4F 56 4D QEMU QXL VGA.OVM 000C0040 46 20 49 6E 74 31 30 68 20 28 66 61 6B 65 29 00 F Int10h (fake). 000C0050 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................ 000C0060 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................UefiSeven:
000C0000 56 45 53 41 00 03 22 00 00 C0 01 00 00 00 2C 00 VESA..".......,. 000C0010 00 C0 30 00 00 00 30 00 00 C0 3A 00 00 C0 47 00 ..0...0...:...G. 000C0020 00 C0 55 65 66 69 53 65 76 65 6E 00 F1 00 FF FF ..UefiSeven..... 000C0030 55 65 66 69 53 65 76 65 6E 00 45 6D 75 6C 61 74 UefiSeven.Emulat 000C0040 65 64 20 56 47 41 00 4F 56 4D 46 20 49 6E 74 31 ed VGA.OVMF Int1 000C0050 30 68 20 28 66 61 6B 65 29 00 90 90 90 90 90 90 0h (fake)....... 000C0060 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 90 ................
It has "text mode", but I wouldn't call it proper.
Most stuff that requires a proper text mode doesn't work with it - only very simple stuff like the COMMAND.COM prompt and very simple text programs work with it, but anything more complex that requires proper text mode will not work - from EDIT, to SCANDISK and so on and so forth.
The only real way to do proper text mode is to use a compatible GPU and the GPU's VBIOS as per title of this issue.
That isn't to say that a fallback is bad to have, but the fallback of SeaVGABIOS as it stands right now is pretty much the best you're gonna get.
@FlyGoat Have you seen OpenCore's CSM implementation ? https://github.com/acidanthera/OpenCorePkg/pull/482
Instead of just using EFI_LEGACY_BIOS_PROTOCOL (which requires CSM support) like rEFInd they implemented all those calls themselves, probably because some Apple devices don't have it and they re-used that code for PC.
It only uses EFI_LEGACY_8259_PROTOCOL, though recent TianoCore verions (and I assume Arrow Lake) are missing that protocol and will need it implemented.
There's also a minimal VGA BIOS implementation (BiosVideo.c) which might be interesting.
https://www.intel.com/content/dam/www/public/us/en/documents/reference-guides/efi-compatibility-support-module-specification-v098.pdf might be of use too.
EDIT:
Weirdly I cannot find a single reference to ExitBootServices within it though, seems like they do that stuff manually instead for some reason. https://github.com/acidanthera/OpenCorePkg/blob/master/Platform/OpenLegacyBoot/LegacyBootSupport.c#L367
@FlyGoat Have you seen OpenCore's CSM implementation ? acidanthera/OpenCorePkg#482
Thanks for the pointer! I actually checked that a couple of days ago.
Instead of just using
EFI_LEGACY_BIOS_PROTOCOL(which requires CSM support) like rEFInd they implemented all those calls themselves, probably because some Apple devices don't have it and they re-used that code for PC.
It's mostly the same as "Skylake+ PAM" fallback unlocking method implemented by CSMWrap.
It only uses
EFI_LEGACY_8259_PROTOCOL, though recent TianoCore verions (and I assume Arrow Lake) are missing that protocol and will need it implemented.There's also a minimal VGA BIOS implementation (BiosVideo.c) which might be interesting.
It also comes from EDK2, but I think this is not a VGA BIOS implementation? It somewhat tried to provide as GOP driver based on VESA VBIOS.
https://www.intel.com/content/dam/www/public/us/en/documents/reference-guides/efi-compatibility-support-module-specification-v098.pdf might be of use too.
EDIT:
Weirdly I cannot find a single reference to
ExitBootServiceswithin it though, seems like they do that stuff manually instead for some reason. https://github.com/acidanthera/OpenCorePkg/blob/master/Platform/OpenLegacyBoot/LegacyBootSupport.c#L367
Just for recording, this is actual EDK2's legacy VBIOS handling code:
https://github.com/tianocore/edk2/blob/stable/202011/OvmfPkg/Csm/LegacyBiosDxe/LegacyPci.c
I just implemented native VGA OpROM support, seems working on RX550, please check current main branch.
Thanks
That's amazing, thank you so much! I'll definitely run some tests now and let you know.
I just implemented native VGA OpROM support, seems working on RX550, please check current main branch.
Thanks
Is this solve VGA problem on GPUs which does not support Legacy OPROM?
I just implemented native VGA OpROM support, seems working on RX550, please check current main branch. Thanks
Is this solve VGA problem on GPUs which does not support Legacy OPROM?
Nah, for such GPU it should be handled by SeaVGABIOS.
I tested on my ARL system with a GPU with a proper OpROM. CSMWrap seemed to say all was good with regards to that, until it got stuck at CALL16 f000:cda5 as mentioned in issue #16.
I tested on my ARL system with a GPU with a proper OpROM. CSMWrap seemed to say all was good with regards to that, until it got stuck at
CALL16 f000:cda5as mentioned in issue #16.
My ARL system (Z890，H810 ) can not disable the above 4G MMIO function. Especially use AMD graphics cards 400s above.
Neither can mine, as long as I have been able to find...
But is that an issue when using an older dedicated graphics card? Or is one forced to use an even older graphics card? Because I haven't tried with the oldest PCIe GPU I have yet...
Neither can mine, as long as I have been able to find...
But is that an issue when using an older dedicated graphics card? Or is one forced to use an even older graphics card? Because I haven't tried with the oldest PCIe GPU I have yet...
I tested the Lenovo machines Q670 with 12-14th cpus, after disable above 4G Decoding and MMIO, they can work perfectly use csmwrap ver 1.1.0, but Z890/H810 can not. I tried to disable it use amisce and ru.efi,even use mod bios, but failed.
I just refreshed seabios build to disable debug prints over serial, which seems causing problems.
If it's still not working, I'll look into getting debug prints via XHCI DbC.
Tested again... No luck.
I even tried the aforementioned oldest PCIe graphics card I have, but that one didn't even get past POST (I assume because it needs a CSM-supporting firmware in order for modeset to be done).
I tried 2 different AMD GPUs on top of that, of different generations; one being an RX 550, and the other a HD6450 - no luck with either of them, it still gets stuck at the CALL16, albeit with a different offset now, but I assume that's due to the different SeaBIOS and matters little.
Oh yeah, I forgot to mention that this machine has a COM port and I am willing and able to test using COM output, instead of going the XHCI route, if need be.
Oh yeah, I forgot to mention that this machine has a COM port and I am willing and able to test using COM output, instead of going the XHCI route, if need be.
It would be nice if you can get logs from serial port :-) Any build prior to 27efbe7b6955a6a47d46579bdbb94f270b6355e1 will print logs on COM1.
Hopefully it's just some SeaBIOS glitch, debugging OpROM is really hard without proper tooling. Maybe it's possible to do it with VFIO.
At least that is without changing CSMWrap's code to address the COM port directly, which I'll likely attempt now.
@Canonkong Legacy VBIOS should still work with 4G Decoding on, but at 640x480 and low bit depth because SVGA modes using VESA require a 32-bit framebuffer address retrieved through INT10H VESA calls, which cannot be passed because it's at a 64-bit address.
The legacy VGA modes with limited resolution/depth are at 0xA0000 which should work as long a legacy VGA decoding is enabled which to my knowledge csmwrap does try to. The chipset might play a role in this though even if the GPU supports it, which could be why it doesn't work on Arrow Lake.
I tested this myself with non-GOP VBIOS on RX 580 but it was on a platform where the chipset supports CSM.
Also FYI any 32-bit OS without PAE that tries to use PCIe devices will have issues with 4G decoding on. It is possible to re allocate and downsize BARs after boot though so it can be worked around on boards where 4G decoding is forced on.
I think we are not far from success.
Testing on Q670 with intel 13th, above 4G MMIO disable, need to wait 1-2minutes，than it can work.
I could not seem to get any output out of the ARL board's COM port... I am so disappointed with this platform, it's just so badly put together, insane Intel released this.
I'll try updating the BIOS to see if that makes any difference, but I'm doubtful it's gonna do anything.
I could not seem to get any output out of the ARL board's COM port... I am so disappointed with this platform, it's just so badly put together, insane Intel released this.
I'll try updating the BIOS to see if that makes any difference, but I'm doubtful it's gonna do anything.
Can we use 1394 cards to debug it？I often use 1394 to make acpi.sys debug.
Testing on Q670 with intel 13th and Nvidia GT730, above 4G MMIO Enable, need to wait 1-2minutes, it also can work.
But can’t set the screen resolution, after install Nvidia graphics cards driver, all can work fine.
I just implemented native VGA OpROM support, seems working on RX550, please check current main branch.
Binary csmwrap.efi link please.
Tested this build: https://github.com/FlyGoat/csmwrap/actions/runs/15122472115
Memory Remap - Enabled
iGPU - SeaVGABIOS
PCIe AMD6450 - AMD VBIOS
Windows 8 Developer Preview 32-bit:
mini XP 32-bit from HBCD14:
WinXP SP2 32-bit + Longhorn 5472 winload.exe loader stuck on aurora bootscreen but works (AMD6450 VBIOS) - Remote Desktop screenshot:
Same WinXP SP2 but loader ntldr+NTDETECT.COM > same error as miniXP HBCD14:
Windows NT has found only 0K of low memory. 512k of low memory
is required to run Windows NT. You may need to upgrade your
computer or run a configuration program provided by the manufacturer.