Re: [Xen-devel] [PATCH 08/11] xen p2m: transparently change the p2m mappings in the m2p override

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

On Thu, Dec 16, 2010 at 03:25:14PM +0000, Stefano Stabellini wrote:
> On Wed, 15 Dec 2010, Jeremy Fitzhardinge wrote:
> > On 12/15/2010 05:40 AM, stefano.stabellini@eu.citrix.com wrote:
> > > From: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> > >
> > > In m2p_add_override store the original mfn into page->index and then
> > > change the p2m mapping, setting mfns as FOREIGN_FRAME.
> > >
> > > In m2p_remove_override restore the original mapping.
> > >
> > > Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> > > ---
> > >  arch/x86/xen/p2m.c |    4 ++++
> > >  1 files changed, 4 insertions(+), 0 deletions(-)
> > >
> > > diff --git a/arch/x86/xen/p2m.c b/arch/x86/xen/p2m.c
> > > index 8db19d5..7dde8e4 100644
> > > --- a/arch/x86/xen/p2m.c
> > > +++ b/arch/x86/xen/p2m.c
> > > @@ -407,8 +407,11 @@ static unsigned long mfn_hash(unsigned long mfn)
> > >  void m2p_add_override(unsigned long mfn, struct page *page)
> > >  {
> > >  	unsigned long flags;
> > > +	unsigned long pfn = page_to_pfn(page);
> > >  	page->private = mfn;
> > > +	page->index = pfn_to_mfn(pfn);
> > 
> > Is there any risk that a page being used for a granted mfn will end up
> > getting xen_create_contiguous_region() applied to it via the DMA API? 
> > That would be messy...
> 
> I don't think so because AFAIK xen_create_contiguous_region is only
> called:
> 
> - by xen_swiotlb_fixup on xen_io_tlb_start (+ offset) that has been
> allocated using alloc_bootmem;

Right, so during bootime.
> 
> - by xen_swiotlb_alloc_coherent on memory allocated using
>   __get_free_pages.

Which is called by the DMA API by 'dma_alloc_coherent' or by 
the PCI API 'pci_alloc_consistent'. So most (if not all) are
done during startup of the device drivers. Thought it
might be that some drivers do allocate it during runtime.

So I think it is OK as long as you get to __set_phys_to_machine
_before_ pci_free_consistent is called.

> 
> if in the future xen_create_contiguous_region will be called on other
> memory ranges then maybe, but at the moment there are no problems.