RE: [PATCH kernel v2 4/4] powerpc/pseries/svm: Allow IOMMU to work in SVM

[Ram Pai <linuxram@us.ibm.com>]

On Fri, Jan 03, 2020 at 11:08:49AM +1100, David Gibson wrote:
> On Thu, Jan 02, 2020 at 02:21:06PM -0800, Ram Pai wrote:
> > On Mon, Dec 16, 2019 at 03:19:24PM +1100, Alexey Kardashevskiy wrote:
> > > H_PUT_TCE_INDIRECT uses a shared page to send up to 512 TCE to
> > > a hypervisor in a single hypercall.
> > 
> > Actually H_PUT_TCE_INDIRECT never used shared page.  It would have
> > used shared pages if the 'shared-page' solution was accepted. :)
> 
> Well, it depends what you mean by "shared".  In the non-PEF case we do
> use a shared page in the sense that it is accessed by both guest and
> hypervisor.  It's just not shared in the PEF sense.

Ah..I see. That sentence can be right or wrong based on the reader's
interpretion of the phrase 'shared page'.  To me a 'shared page' is a
page that is **explicitly** shared with the hypervisor. However I can see
'shared page' to mean a page that is simply shared; either implicitly or
explicitly.

Given that, there is a possibility for mis-interpretation, I think, it
might be better to avoid the phrase 'shared page' if possible.

> 
> > > This does not work for secure VMs
> > > as the page needs to be shared or the VM should use H_PUT_TCE instead.
> > 
> > Maybe you should say something like this.. ?
> > 
> > H_PUT_TCE_INDIRECT does not work for secure VMs, since the page
> > containing the TCE entries is not accessible to the hypervisor.
> > 
> > > 

..snip.

-- 
Ram Pai