[LinuxBIOS] booting XP

[ron minnich]

The issue of booting XP just will not go away.

So, there are two options I see, in order of preference:
1. find some way, using kexec etc., to boot XP under Linux, then put Linux
in FLASH; done
2. revive ADLO

there have been a number of suggestions for how to do the first one.
1. freeldr (vapor?)
2. tinyldr (vapor?)

any non-vaporous ones out there? This note is an attempt to start a
discussion and discovery on this subject.

thanks

ron

[Stefan Reinauer]

* ron minnich  [061107 04:25]:
> So, there are two options I see, in order of preference:
> find some way, using kexec etc., to boot XP under Linux, then put Linux in
>
> FLASH; done
> 2. revive ADLO

Maybe be can join 1 and 2, and run (parts of) ADLO on top of Linux, 
together with kexec.
> there have been a number of suggestions for how to do the first one.
> freeldr (vapor?)
> tinyldr (vapor?)

Any of these might be obsolete with Windows Vista, as it uses
"winload.exe" instead of NTLDR

The ugly thing with all Windows booting approaches is that the boot
loader has a whole lot of intelligence, it is doing device enumeration,
and, a lot worse: It is also loading device drivers. :-(

I found a little bit of information about the boot configuration data
provided by the boot loader to NTOSKERNL.EXE (Vista version)

http://download.microsoft.com/download/a/f/7/af7777e5-7dcd-4800-8a0a-b183365...

Microsoft also has an online reference:
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/BCD/bcd/por...

Also there's nice information about the XP boot process:
http://www.geocities.com/asoke_dasgupta/boot-xp.html

Wikipedia knows:
http://en.wikipedia.org/wiki/Windows_XP_Startup_Process

[Stefan Reinauer]

* ron minnich  [061107 04:25]:
> The issue of booting XP just will not go away.
>
> So, there are two options I see, in order of preference:
> find some way, using kexec etc., to boot XP under Linux, then put Linux in
>
> FLASH; done
> 2. revive ADLO
>
> there have been a number of suggestions for how to do the first one.
> freeldr (vapor?)
> tinyldr (vapor?)

I had a quick chat about tinyldr this morning with Alex Ionescu, the
author of tinykernl.

It seems we can actually use most of NTLDR without any change.
The only component we actually have to provide seems to be "startup.com"
which is an abstracted interface doing all the bios callbacks.

I consider this really great news.

He sent me this URL:
http://svn.tinykrnl.org/svn/tinykrnl/base/boot/tbx86/

TBX86 is his version of startup.com.

Stefan

[Segher Boessenkool]

> It seems we can actually use most of NTLDR without any change.
> The only component we actually have to provide seems to be  
> "startup.com"
> which is an abstracted interface doing all the bios callbacks.
>
> I consider this really great news.

If you care about mswindows at all, it sure is.  Does this work
for vista too though?

Segher

[Stefan Reinauer]

* Segher Boessenkool  [061107 10:16]:
> If you care about mswindows at all, it sure is.

according to OSnews.com as well as
http://www.w3schools.com/browsers/browsers_stats.asp Windows has a
market share of about 90% in the PC market.

If LinuxBIOS is at all aiming at becoming an alternative firmware 
for PCs, really nobody would assume we can afford not to care.

Given that we can only keep up with supporting new server chipsets
if we have a user base worth mentioning, there is no way around
targeting the PC market as well.
> Does this work for vista too though?

We will need to find out by trying. There is no single future proof path
or optimal solution unfortunately.

- The "16bit bios callback approach"

Using ADLO or something similar will help for Windows Vista, but it
  would need work on the built-in IDE driver, plus it would need work to
  get int13 plugin cards working.

The strength of this approach is that by using a 16bit legacy
  interface cleanly encapsulated, we might be able to boot from any
  PCI plugin card, such as SCSI adapters.

The weakness is clearly that we need to stick with 16bit legacy code
  (which we already do for vga init anyways)

It sounds somewhat weird that we need to go ancient to support 
  newer systems than what is there today, but this might indeed be the
  case.

- The patchwork approach

We create clean 32bit loaders for the operating systems we want to
  boot. Or we modify or enhance existing loaders to do so.

The strength of this approach is we might be able to use Linux for
  booting from flash, gaining from Linux' good driver support.

The weakness of this approach is we need to reflash the BIOS if we
  plug in a new SCSI adapter because the old one died. All SCSI drivers
  in Linux require about 5MB space, 1.3 for all ide devices, 1.5 for
  Infiniband. Even with pretty good compression, how are we going to
  plug this, plus a kernel, plus a bootloader plus LinuxBIOS in 1MB
  flash?

Even though I am sure almost all people on this list compile their own
  kernels, this is not the usual case for people using Linux anymore.
  _Requiring_ to recompile your _BIOS_ to support booting from a new PCI 
  card will not make us a lot of friends.

So here we go. That's quite some of the misery along the mainstream
path. Solutions, anyone?

[ron minnich]

for now, let's get adlo back to the point that we can use it? That includes
documentation, and possibly setting up a buildrom target for it.

ron

[ron minnich]

wow, looking at ADLO, this will be a bit of work, but worth it.

What's the accepted way to bulid 16-bit code nowadays? Still bcc?

as86 loader.s is easy: convert and junk as86.

I assume that vista will boot on legacy pc with 16-bit bios, right?

ron

[Stefan Reinauer]

* ron minnich  [061107 16:47]:
> wow, looking at ADLO, this will be a bit of work, but worth it.
>
> What's the accepted way to bulid 16-bit code nowadays? Still bcc?
>
> as86 loader.s is easy: convert and junk as86.
>
> I assume that vista will boot on legacy pc with 16-bit bios, right?

and only with that. They refused to put EFI support in there. Funny, but
Microsoft is a little bit on our side ;-)

[ron minnich]

Challenge to the list: find that port patch Richard mentioned. I have to go
to a meeting. If someone can find it, I'll check it out and commit it.

ron

[Stefan Reinauer]

* ron minnich  [061107 16:53]:
> Challenge to the list: find that port patch Richard mentioned. I have to go to
> a meeting. If someone can find it, I'll check it out and commit it.

The full source is here (sent around by YhLu)
LinuxBIOSv2/util/ADLO

Added in r2457 (found using the search engine on qa.linuxbios.org)

http://snapshots.linuxbios.org/stats/commit-LinuxBIOSv2-2457.log

[Richard Smith]

Stefan Reinauer wrote:
> ron minnich  [061107 16:53]:
> > Challenge to the list: find that port patch Richard mentioned. I have to go to
> > a meeting. If someone can find it, I'll check it out and commit it.
>
> The full source is here (sent around by YhLu)
> LinuxBIOSv2/util/ADLO
>
> Added in r2457 (found using the search engine on qa.linuxbios.org)

Now. that I did miss..

[yhlu]

I have one idea about.
I assume all new system will be dual way or dual cores.

So I let first core run at 16 bit to provide 16 bit interface, and
second core runing as 32 bit with Linux kernel as service.

YH

On 11/7/06, Stefan Reinauer  wrote:
> ron minnich  [061107 16:53]:
> > Challenge to the list: find that port patch Richard mentioned. I have to go to
> > a meeting. If someone can find it, I'll check it out and commit it.
>
> The full source is here (sent around by YhLu)
> LinuxBIOSv2/util/ADLO
>
> Added in r2457 (found using the search engine on qa.linuxbios.org)
>
> http://snapshots.linuxbios.org/stats/commit-LinuxBIOSv2-2457.log

[Stefan Reinauer]

* yhlu  [061107 17:43]:
> I have one idea about.
> I assume all new system will be dual way or dual cores.
>
> So I let first core run at 16 bit to provide 16 bit interface, and
> second core runing as 32 bit with Linux kernel as service.

Asynchronous Multi Processing? :-)

[Richard Smith]

Here's a novel idea:

How about we ask the original author Adam Sulmicki 
what he thinks needs to be done to boot XP?

I believe he still reads the list but I've cc'ed him just in case.

[Johan Rydberg]

Stefan Reinauer  writes:
> > What's the accepted way to bulid 16-bit code nowadays? Still bcc?
> >
> > as86 loader.s is easy: convert and junk as86.
> >
> > I assume that vista will boot on legacy pc with 16-bit bios, right?
>
> and only with that. They refused to put EFI support in there. Funny, but
> Microsoft is a little bit on our side ;-)

There will be EFI support for 64-bit hostss, IIRC.  They just rejected
it for 32-bit hosts.

~j

[Stefan Reinauer]

* Johan Rydberg  [061107 17:47]:
> There will be EFI support for 64-bit hostss, IIRC.  They just rejected
> it for 32-bit hosts.

So we can run Windows on 64bit using GNUFI soon? ;-)