User-mode Linux kernel port

On Fri, Apr 25, 2008 at 04:02:26PM +0200, Daniel Janzon wrote:
> Are there any reasons UML will not be able to run on for instance ARM
> platforms? 

Nope.

> I see in arch/um that there is some processor-dependent code
> for x86 and ppc. Is this code strictly needed in order to run UML? Or is
> only for added performance?

A bunch of architecture-specific stuff shows through the system call
interface, and the arch/um/sys-* (and related headers) exist to
abstract that stuff away from the rest of UML.  So, you will need to
implement the relevant stuff for ARM if you want UML/arm.

	      	       	     Jeff

-- 
Work email - jdike at linux dot intel dot com

On Mon, 2008-04-28 at 12:47 -0400, Jeff Dike wrote:
> On Fri, Apr 25, 2008 at 04:02:26PM +0200, Daniel Janzon wrote:
> > Are there any reasons UML will not be able to run on for instance ARM
> > platforms? 
> 
> Nope.

Great!

> > I see in arch/um that there is some processor-dependent code
> > for x86 and ppc. Is this code strictly needed in order to run UML? Or is
> > only for added performance?
> 
> A bunch of architecture-specific stuff shows through the system call
> interface, and the arch/um/sys-* (and related headers) exist to
> abstract that stuff away from the rest of UML.  So, you will need to
> implement the relevant stuff for ARM if you want UML/arm.

Thanks for the information. Two follow-up questions though. First, are
there any certain features of the kernel that can be turned off that
makes the porting effort substantially smaller? And secondly, can you
recommend any documentation that will set me off in the right direction?
Otherwise the source is my friend, I guess :)

       Best regards,
       Daniel

On Tue, Apr 29, 2008 at 10:27:54AM +0200, Daniel Janzon wrote:
> Thanks for the information. Two follow-up questions though. First, are
> there any certain features of the kernel that can be turned off that
> makes the porting effort substantially smaller? 

Turn off hostfs.

You can make the smallest possible kernel that you can boot, but
it doesn't seem to me that much else will make a big difference in the
porting effort.

> And secondly, can you
> recommend any documentation that will set me off in the right direction?
> Otherwise the source is my friend, I guess :)

There's http://user-mode-linux.sourceforge.net/old/arch-port.html, but
it's horribly out of date.  The issues described there are obviously
the same, but the interfaces described there are different now for the
most part.

The arch-specific stuff is in
    include/asm-um/*-$(ARCH).h
    arch/um/sys-$(ARCH)
    arch/um/os-Linux/sys-$(ARCH)
    arch/um/include/sysdep-$(ARCH)

Plus arch/um/Makefile-$(ARCH) and arch/um/Kconfig-$(ARCH).

			Jeff

-- 
Work email - jdike at linux dot intel dot com

Le Mardi, 29 Avril 2008 11:21:29 -0400,
Jeff Dike <jdike@...> a écrit :

> You can make the smallest possible kernel that you can boot, but
> it doesn't seem to me that much else will make a big difference in the
> porting effort.

I noticed this thread.  By the subject it seems it's taken for granted
that UML will run on a PowerPC.  is this the case ?

Al

On Thu, 2008-05-01 at 18:57 -0400, lanas wrote:
> Le Mardi, 29 Avril 2008 11:21:29 -0400,
> Jeff Dike <jdike@...> a écrit :
> 
> > You can make the smallest possible kernel that you can boot, but
> > it doesn't seem to me that much else will make a big difference in the
> > porting effort.
> 
> I noticed this thread.  By the subject it seems it's taken for granted
> that UML will run on a PowerPC.  is this the case ?
> 
> Al


The only reason I assumed there was PowerPC support was that there is a
directory sys-ppc (as well as sys-i386, sys-ia64 and sys-x86_64)
in /arch/um in the kernel source. I have idea about the condition of
those files; they may be age-old attempts for PowerPC-support only.

BTW, is there any difference between PowerPC and PPC? In /arch in the
kernel source, there is both a powerpc and a ppc directory. This has
always puzzled me. (I do realize it's a bit off-topic, sorry for that.)

      Daniel

Hi,

On Mon, May 05, 2008 at 01:48:36PM +0200, Daniel Janzon wrote:
> On Thu, 2008-05-01 at 18:57 -0400, lanas wrote:
> > I noticed this thread.  By the subject it seems it's taken for granted
> > that UML will run on a PowerPC.  is this the case ?
> 
> The only reason I assumed there was PowerPC support was that there is a
> directory sys-ppc (as well as sys-i386, sys-ia64 and sys-x86_64)
> in /arch/um in the kernel source. I have idea about the condition of
> those files; they may be age-old attempts for PowerPC-support only.

It was pretty much working once (many years ago, around 2.4.0ish, I
think), but is way out of date now (eg this was before the SKAS patches,
when ptrace was the only option).

Regards,

Chris