Re: [Qemu-devel] [qemu PATCH v2 3/4] nvdimm, acpi: support NFIT platform capabilities

[Ross Zwisler]

On Fri, May 18, 2018 at 04:37:10PM +0000, Elliott, Robert (Persistent Memory) 
wrote:
> 
> 
> ...
> > Would it help to show them in hex?
> > 
> >   As of ACPI 6.2 Errata A, the following values are valid for the bottom
> >   two bits:
> > 
> >   0x2 - Memory Controller Flush to NVDIMM Durability on Power Loss Capable.
> >   0x3 - CPU Cache Flush to NVDIMM Durability on Power Loss Capable.
> 
> Yes, that helps (unless the parser for that command-line does not 
> accept hex values).
> 
> It would also help to make the text be:
>       "CPU Cache and Memory Controller Flush"

Yea, let me check on that.  I'll update the wording regardless to try and make
it more clear.

> ...
> > > So, there should be a way to specify a highest_cap value to convey that
> > > some of the upper capabilities bits are valid and contain 0.
> > 
> > Right, I'll make this dynamic based on the capabilities value passed in by
> > the user.  That's a much better solution, thanks.  This should cover all the
> > same cases as you have outlined above, without burdening the user with yet
> > another input value.
> 
> Automatically determining the highest bit that the user wants to set to 1
> should be easy, and will probably be the most common case.
> 
> It's harder to let the user set some upper bits to 0 but also have them
> within the highest_cap range.  Since this will be less common, the syntax
> could be more convoluted, like an optional highest_cap argument
> to override the automatically generated value.
> 
> For example, to report bits 7, 1 and 0 are all set to 1:
>       -machine pc,accel=kvm,nvdimm,nvdimm-cap=0x83
> would automatically set highest_cap to 7.
> 
> To report bit 7 set to 0 while bits 1 and 0 are set to 1:
>       -machine pc,accel=kvm,nvdimm,nvdimm-cap=0x3,nvdimm-highest-cap=7

Yea, I agree that this is how we could do it, but I don't think this is
necessary right now.  We currently only have 3 bits in the Capabilities field,
and right now there is never a case where there is a difference between "I
don't know about this bit" and "I know about this bit, and it's value is 0".

So, really for now we could essentially just say the highest_cap = 31 and be
fine.

Let's put off the nvdimm-highest-cap argument complexity until we actually
have a use case where it adds value.