Re: [Qemu-devel] [PATCH for-2.9 v2 2/2] intel_iommu: provide "aw-bits" parameter

[Alex Williamson]

On Tue, 13 Dec 2016 15:25:03 +0800
Peter Xu <address@hidden> wrote:

> Previously vt-d codes only supports 39 bits iova address width. This
> patch provide a new parameter for Intel IOMMU to extend its address
> width to 48 bits.
> 
> After enabling larger address width (48), we should be able to map
> larger iova addresses in the guest. To check whether 48 bits aw is
> enabled, we can grep in the guest dmesg with line: "dmar: Host address
> width 48".
> 
> To keep compatibility with old QEMU and most hosts, default aw bits are
> set to 39 bits.

I really fail to see the benefit of this option.  Clearly we always
need to be able to support identity maps through the IOMMU, so the
address width of the vIOMMU needs to support the maximum address of any
memory or mmio range available to the VM.  However if that's the case,
then the address width of the vIOMMU should be automatically configured
for the VM configuration.  What's the benefit of creating yet another
obscure device option rather than configuring this automatically.  If
the argument is to be able to test 48bit vIOMMU support, then we could
simply prefix this option with 'x-' to indicate an experimental,
unsupported option.  Creating this option with the expectation that the
user must specify this in order to potentially create a valid VT-d
configuration for the VM doesn't make sense to me.  Thanks,

Alex