PAE status

PAE status

[Gerd Knorr]

  Hi,

I've just uploaded a fresh batch of patches (#10) for
unstable to http://dl.bytesex.org/patches/

current state:
 * xen itself boots with PAE paging enabled.
 * linux in domain0 boots.
 * hypercall interface has been changed: updates are passed
   via pfn & flags in separate variables.  WARNING: this
   breaks backward compatibility.
 * Using up to 16 GB should work now in theory.

known issues:
 * Using memory above 4 GB is completely untested.  I wouldn't
   be surprised if I missed some places which need fixing for
   36-bit physical addresses ...
 * The patches break the x86_64 build.

Testing, patch reviews & comments are welcome.
Below is a description of the individual patches.
The "series" file lists the apply order of the
patches.

Enjoy,

  Gerd

xen patches
===========

pae-support
  groundwork for PAE support.

hypercalls
  updates the hypercall interface to pass pgtable entries
  using pfn & flags separated.

pae-enable
  enables PAE in the build.

pae-memlimit
  raise memory limit to 16 GB.


linux-xen patches
=================

linux-2.6.11.10.patch
x86_64-linux.patch
iomap.patch
agpgart.patch
i386-cpu-hotplug-updated-for-mm.patch
  stuff from the patches directory

shared-pmd
  Add a #define for shared kernel pmd (should go into the patches
  subdir, also should be pushed upstream ...).

arch-xen-unstable
  auto-generated patch which adds arch/xen to a vanilla kernel source
  tree.

xen-early-printk
  hackish early printk, commented out at the moment so it has no effect,
  but might be useful for debugging.

pae-support
  PAE support groundwork.

vmalloc-size
  make the size of the vmalloc area depend on the domain type, i.e. make
  it much smaller for unpriviliged domains which don't need address
  space for ioremap() & vmalloc().

hypercalls
  updates the hypercall interface to pass pgtable entries
  using pfn & flags separated.

Re: PAE status

[Gerd Knorr]

Gerd Knorr <kraxel@suse.de> writes:

> known issues:
>  * Using memory above 4 GB is completely untested.  I wouldn't
>    be surprised if I missed some places which need fixing for
>    36-bit physical addresses ...
>  * The patches break the x86_64 build.

Found a bug while cleaning up stuff a bit to make it build again on
x86_64.  To fix that I need 64bit support for i386 in cmpxchg(), can
someone (Keir?) please add this?

thanks,

  Gerd

Re: PAE status

[Keir Fraser]

On 25 May 2005, at 10:42, Gerd Knorr wrote:

> Found a bug while cleaning up stuff a bit to make it build again on
> x86_64.  To fix that I need 64bit support for i386 in cmpxchg(), can
> someone (Keir?) please add this?

Okay, I'll add it this afternoon.

  -- Keir

Re: PAE status

[Gerd Knorr]

Keir Fraser <Keir.Fraser@cl.cam.ac.uk> writes:

> On 25 May 2005, at 10:42, Gerd Knorr wrote:
> 
> > Found a bug while cleaning up stuff a bit to make it build again on
> > x86_64.  To fix that I need 64bit support for i386 in cmpxchg(), can
> > someone (Keir?) please add this?
> 
> Okay, I'll add it this afternoon.

Thanks, uploaded a new set of patches (#11) to
http://dl.bytesex.org/patches/

Has this bug fixed (so there is a chance it actually works with more
than 4 GB now).  It also doesn't break the x86_64 build any more.

  Gerd

-- 
-mm seems unusually stable at present.
	-- akpm about 2.6.12-rc3-mm3