Tyan serial console - how?
bari at onelabs.com
Thu Nov 7 21:06:15 PST 2002
Maurice Hilarius wrote:
> There is the LinuxBIOS project, at LANL.
> I am sure they would not object to somebody contributing/helping.
> Of course we have to talk to Award and Phoenix ( really the same
> company, BTW) to give us code to start with..
> And that is the REAL problem.
> It's all deep dark secrets.
> Why else do you think most BIOS suck to some degree?
> All these guys are ever doing is scrambling to make and fix BIOS for
> the latest and greatest boards and chipsets. With VIA, Intel, and so
> on making so darned many chipsets nowadays I am amazed they can keep
> up at all.
> For example the APIC code for the AMD 766/768 chips was so bad that
> Tyan finally gave up on the Phoenix part of that and wrote their own
> to get it done properly.
> I know, as we built a number of machines this spring that needed a PCI
> riser with an active PCI bridge, and a DLink DFE578 with it's own PCI
> The Phoenix based version of the BIOS could not handle giving that
> mess IRQs to save it's life.
> Tyan did their own code to fix it.
> BTW, good luck in trying to get Intel to fix a BIOS problem.
> Unless you are Dell, IBM, or HP it probably won't happen.
> We ran into this on BX boards a few years ago, and realized that the
> last board you want to use on a cluster is an Intel. If you flashed
> those up to newer versions they did not let you flash back down again
> successfully. Ended up finding an old one and using an EPROM burner to
> recover a cluster after a customer upgraded his BIOS on the nodes.
> Intels response was essentially a blank stare when we asked them for
LinuxBIOS support for the AMD 760 has been up for about a year. The port
went public back in August after AMD finally opened up the specs to get
the cache working.
The push now is to have LinuxBIOS up for all the embedded and server
type chipsets around the same time the chipsets are released to the public.
You can also get the source from Award/Phoenix, but it's not exactly
free or inexpensive ( >$50K). Then hack away same as they do at getting
their old legacy spaghetti code to work to roll your BIOS.
More information about the Beowulf