On Tue, 2006-08-22 at 09:42 +0200, Tristan Gingold wrote:
> Le Lundi 21 Août 2006 21:07, Hollis Blanchard a écrit :
> > On Mon, 2006-08-21 at 17:18 +0200, Tristan Gingold wrote:
> > > I am posting the linux xencomm code for review.  I'd plan to submit soon
> > > unless comments/remarks.
> >
> > NAK. I'm still waiting to hear back about how you can use
> > xencomm_inline() without worrying about page boundaries.
> More elaborated answer:
> 
> On linux/ia64, kernel is linearly mapped into guest physical memory.  The 
> same 
> is true for process kernel stacks.  Therefore all kernels structure are 
> linear in guest physical memory.
> 
> Kernel data may of course cross page boundaries.  But Xen can correctly 
> handle 
> this using only the guest physical address.

I see, and you handle this by breaking up copies at page granularity in
xencomm_copy_from_user().

I have to say I don't really like the code complexity, or the fact that
there are now two very different ways to access guest handles. That
being said, it sure would be nice to get rid of that "mini" stack-based
stuff, so it's OK with me. I would be surprised if there were any
performance difference, however.

I'll send some comments to your original patch.

-- 
Hollis Blanchard
IBM Linux Technology Center


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

Reply via email to