Archived from groups: comp.sys.ibm.pc.hardware.chips,comp.arch (
More info?)
"Robert Myers" <rmyers1400@comcast.net> wrote in message
news:1tcs21ltapnsp1fdb6jn8lk0h1vvuffkvp@4ax.com...
> On Tue, 08 Mar 2005 23:31:40 GMT, "FredK" <fred.nospam@nospam.dec.com>
> wrote:
>
> >
> >"Robert Myers" <rmyers1400@comcast.net> wrote in message
> >
> >snip
> >
> >> >
> >> Intel wanted to provide a standard way to boot IA-64. A standard PC
> >> boot couldn't be used because itanium could not start (or execute,
> >> even) in IA-32 real mode. They had to do something, and EFI is what
> >> they did.
> >>
> >> I haven't had much of a chance to play with EFI, but what I've seen
> >> (you can get in there between the firmware and the OS, if you have to,
> >> and EFI applications run on a virtual machine) seems pretty
> >> attractive.
> >>
> >
> >EFI is a small, very simple OS. It's not really a virtual machine
> >environment.
>
> I'm sorry. I didn't mean to imply that the environment itself, which
> as far as I know is linux, runs on a virtual machine. You'd have to
> run a byte code interpreter in that environment to execute the kind of
> EFI code that is interpreted at boot, but the intent, I believe, is
> that device drivers be independent of CPU.
>
Nope. EFI and EFI applications are architecture specific - IA32 or
IA64 right now. The ACPI layer uses an interpreter for the "code"
that knows how to enumerate the platform and devices, and also
how to access that platform specific hardware (like a core chipset).
I also believe that the object format in the PCI devices ROM may
be an interpreted byte stream - but I don't recall offhand.
> <snip>
>
> >>
> >> Sure, the whole OS could migrate into EFI. It's already happened.
> >> With no disk installed, my itanium box boots into linux. Not an
> >> especially powerful linux, but that's a detail. You can boot into
> >> linux, fiddle things if you like, and then boot a disk-based OS.
> >>
> >
> >Well... this isn't true as far as I know. Boot loaders are a special
form
> >of EFI application used to load some OS specific code, transfer control
> >and never return. The Linux method is a EFI application called "elilo"
> >which is normally resident on a hard drive in the FAT formatted EFI
> >partition. "elilo" pulls in an image that contains a Linux kernel and
> >a memory disk. Or you could do a network boot using PXE (bootp on
> >steroids). But I don't know of anyone who is blasting "elilo" or the
boot
> >kernel into flash ROM.
> >
> I'm not sure what I said that you're disagreeing with. As far as I
> know, the prompt I get at boot is linux.
An Itanium system will initialize to a boot menu. One selection is
"shell" -
this is not Linux. If you put a Linux distribution CD into the system, it
"should" auto-boot it into a simple Linux kernel. But Linux isn't in the
ROM AFAIK on any Itanium system.
>
> Sure, the widget that loads Linux (or whatever) loads from disk and
> runs as a garden-variety program, but there is no reason at all why
> full functionality couldn't be in firmware and, I believe, such a
> thing will come to pass for thin, diskless clients. Such a client
> will pull its operating system in via PXE? Maybe. No reason why it
> has to, though.
>
It is certainly possible to blast it into the ROM or Flash instead of
having it disk resident.