The problems is not compatibility, it is a broken model.

Posted May 14, 2007 13:40 UTC (Mon) by brugolsky (guest, #28)
In reply to: BIOS compat issues by arjan
Parent article: Free software drivers for the Intel 965GM Express Chipset

Thanks, I took note of your original announcement, and then it passed off my radar. This and PowerTOP look useful, and I certainly appreciate the efforts of you and others at Intel to improve the situation. So take the following as the rantings of someone driven mad by machine room noise. :-P

As I'm sure that you are aware, if you want to help Linux on Intel boxes, get Intel to participate in LinuxBIOS, and just get us to userland as quickly as possible. After putting the machine in the rack and cold booting, I have almost no use for the BIOS - I want to kexec into my next kernel, and I want the hardware to work properly. If SMI is needed for anything, it should be fixing up my hardware state when I kexec -- other than that, it should get out of the way.

I am currently grappling with trying to force-enable HPET on a number of motherboards -- between the HPET doc and the patches posted to LKML, I'll probably get it working - but what a waste of time, and why? Serial consoles on a large number of motherboards have inconsistent settings (some do 9600, some only do 57600, some on ttyS0, some on ttyS1), have broken terminal handling, and hang in various ways. A hint to BIOS vendors: terminal escape sequences and lack of flow control don't mix well; the whole business of exporting the console display is just braindead. I want a boot log, I want it stored somewhere, and I want to be able to dump it into /var/log/ once my machine is up. Why it can't have a real settable verbosity level. Even GRUB can do better than that. We have Intel-SE7320VP2 motherboards with Marvel and Intel E1000 NICs; of course the machine can't PXE boot from the E1000. Every BIOS seems to have a different PXE enable/disable behavior, and it presents so many headaches in a VLAN and security rich environment that we use USB sticks on our DMZ machines instead.

My brand-new Core2 Duo ICH7-based laptop is currently using ata_piix, for lack of a config option to switch to AHCI. :-(

I went to the EFI sessions at OLS a few years ago, and I was appalled. I apparently still can't do something like:

diff -u <(ssh foo efi-config-dump) <(ssh bar efi-config-dump) | less
ssh foo efi-config-dump | ssh bar efi-config-restore

Perhaps the engineers should have a look at the "lxbios" tool.

Where are the config commands that would allow me to do a "setpci" to workaround a misconfigured bridge? Perhaps it is there now, but I wouldn't know, because I walked away in disgust.

EFI introduces yet another driver model. One would think that nobody at Intel is aware of the problems with MILO, SILO, GRUB, OpenBoot etc. EFI is so bloated, many BIOS flash parts now have enough room for a whole mini Linux distro. I guess I should count that as a blessing; someday LinuxBIOS may run on the handful of new Intel chipsets, and then we'll have plenty of room to install a mini-distro in flash.