Re: [PATCH] libxc: introduce a per architecture scratch pfn for temporary grant mapping

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 14/01/15 13:31, Ian Campbell wrote:
> On Wed, 2015-01-14 at 13:23 +0000, Andrew Cooper wrote:
>>> ARM is defining a grant table range which is out of the RAM and not
>>> mapped at that time. So we can reuse it with any possible issue.
>> Do you mean to say that there is a grant table range baked into the ABI
>> occupying guest physical address space?
> A range of otherwise unused IPA space is suggested to the guest via the
> device tree node, as a convenience for the guest so it doesn't need to
> figure one out for itself (it's morally equivalent to the BAR provided
> on the x86 platform PCI device).
>
> So the actual address is not part of the guest ABI, but the toolstack
> does know what it is going to be since it is privy to some internal
> details of the guest layout.

Ah ok, so you can guarentee to find an area which doesn't have RAM in it.

However, being common code, the actions are still to add to the physmap,
map, edit, unmap and remove from the physmap, which is inefficient.

I still feel that the concept of a scratch gfn is flawed, and that a
mapping hypercall which takes {domid, space, idx} as parameters is the
correct solution.  (Although thinking about, this probably requires a
new privcmd ioctl? urgh)

As this change is a bugfix for ARM and makes the common situation no
worse, it is possibly acceptable on those grounds.

~Andrew