Re: RFC Patch for the "x86-64, mm: Put early page table high"

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

> >>> So I decided to look at this from a different perspective: why not
> >>> contain the workaround inside Xen early bootup code.
> >>>
> >>> I've tested this code as DomU (1G, 2G, 3G), Dom0 (8G, 4G, 1000M)
> >>> and they all booted fine. Hadn't compile tested it on 32-bit and
> >>> I think it will actually complain about it. Let me fix that.
> >>>
> >>> See attached patch (also present in stable/bug-fixes-for-rc5) which also
> >>> has the "xen: mask_rw_pte mark RO all pagetable pages up to pgt_buf_top"
> >>> integrated in.
> >> Well, if we can hide the fix away in our code, then that has obvious
> >> advantages.  But I worry that this change is pretty closely dependent on
> >> how the other code works, and would be fragile in the face of further
> >> changes to that code (esp since there's no obvious coupling between the
> >> two, so anyone changing the arch/x86 code wouldn't have any clues to
> >> look at the corresponding Xen code).
> > True. I am hoping to remove this in 2.6.40 though...
> 
> That makes it more palatable.  What's the solution for 40+?

None yet. I am thinking to schedule some form of meeting between hpa, Yinghai,
Stefano, you, and whoever else to hammer something out.

So far we got two workarounds in the Xen MMU code - there is the 32-bit dance with
the swapper_pg being copied from initial_..something and then back (Ian came up
with a fix for that), and then this one.

Did you have a chance to test this patch on your box?