Re: [Xen-ia64-devel] Keyboard and Mouse are disable in Xen0 terminal

[Isaku Yamahata <yamahata@xxxxxxxxxxxxx>]

Hi Kevin.

On Mon, Jun 19, 2006 at 09:04:37PM +0800, Tian, Kevin wrote:
> Hi, Kangkang,
>       This phenomenon is rooted caused by Rev 10373 which adds 
> mfn_to_local_pfn to xen/ia64. The problem is that this inline function 
> is implemented on top of mfn_to_pfn (dummy passthrough) instead 
> of mfn_to_pfn_for_dma, while the place to call mfn_to_local_pfn is in 
> swiotlb. Not sure why only USB device is affected by this bug.
> 
>       Yes, the quick solution can be to use mfn_to_pfn_for_dma version 
> for now. However this interface may be also used by places without 
> dma requirement where requires the VP concept. Maybe time for 
> considering real p!=m model instead of the VP model in the middle? 
> Considering two set of interfaces existing to accommodate VP and 
> p!=m difference which really confuses people and error prone like this 
> bug. Isaku, how about your opinion?

I agree with you that the current for_dma scheme is error-prone
and confusing.
In fact I was aware of mfn_to_local_pfn(), but I overlooked the bug.
The short term solution is to use mfn_to_pfn_to_for_dma().


For the middle term.
I think this is a good evident to have IA64 specific
swiotlb (and maybe also pci-dma-xen).
Using i386's swiotlb and pci-dma-xen is a temporal hack and
it should be cleaned up.

My view is as follows.
Although dma-paravirtualization is different from mmu-paravirtualization,
xen/x86 confuses them. Ideally it is right to treat them differently.
However I don't think the xen/x86 developper will accept such a change.
Probably things may change when IOMMU support becomes a development issue.

Xenlinux/ia64 has quite different concept about mmu-(para)virtulization
and dma-paravirtulization from xenLinux/x86's (for now).
However swiotlb and pci-dma-xen are shared by them.
The sharing is the root cause, so creating IA64 specific swiotlb and
pci-dma-xen is the solution.
Cleaning up xenLinux/IA64 paravirtualization for Linux upstream merge
is also another middle term goal.
Maybe it means to create a new system type something like mach-xen.
I think these clean up can/should be done at the same time.


It seems that you'd like to change the current Xen/IA64
mmu-fullvirtualization.
Do you have any other motivation? performance?

Thanks
-- 
yamahata

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel