Re: [Qemu-devel] [PATCH 0/5] Extend TPM support with a QEMU-external TPM

[Igor Mammedov]

On Mon, 04 May 2015 14:39:32 -0400
Stefan Berger <address@hidden> wrote:

> On 05/04/2015 12:16 PM, Kevin O'Connor wrote:
> > On Mon, May 04, 2015 at 11:22:25AM -0400, Stefan Berger wrote:
> >> On 05/04/2015 05:16 AM, Igor Mammedov wrote:
> >>> On Wed, 29 Apr 2015 12:42:21 -0400
> >>>> 2 choices now -- which one to take?
> >>> I'd try installing extra SSDT table first as a cleanest way
> >>> (seabios only) and if it fails fallback to TIS path.
> >> I did some experiment where I tried to use AML's Store() call to
> >> see whether it can read from some entry in another scope, similar
> >> to what the specs show on page 487, so that one could read the
> >> base address that is written elsewhere. The problem seems to be
> >> that there is no actual searching going on for accessing a scope
> >> during runtime, but this seems to be done during compile time. So
> >> something like Store(\_SB.I2C1.Y, BUFF) from the specs needs to
> >> have a SCOPE(\_SB.I2C1...) available during compile time.
> > You can reference scopes provided by other SSDTs/DSDT using the
> > "External()" acpi command.
> 
> Tried it. Works with the \DBUG function for example.
> 
> As an example, External(\BASE) would give allow us to get the base
> address. The AML code for BASE, something like this
> 
>      Scope(\) {
>          Name(BASE, 0x12345678)
>      }
> 
> 
> Access to it would be like this:
> 
> Store(\BASE, Local0)
> 
> The above first snippet would have to be compiled by SeaBIOS.
> 0x12345678 would in this case be the address allocated by SeaBIOS
> where the AML code for PPI could read/write the PPID data to/from.
> Kevin and I had an offline conversation about this and Kevin doesn't
> seem to like the idea of a split implementation of ACPI table
> generation. I would then extend the TIS with a couple of bytes that
> retain the data across a reset.
It's not really split implementation, it's installing an additional
table which is purely SeaBIOS specific since QEMU doesn't care about
PPID data at all nor does it need to do so.

For example of AML template patching in SeaBIOS you can look at
build_pcihp(), so you can dynamically patch AML variable in template
when SeaBIOS (re)boots.

What's missing in SeaBIOS is a mechanism to add extra table to QEMU
supplied RSDT.

But I'm fine with paravirt implementation if above is considered too
hard to implement. If you go this route look at build_ssdt() in QEMU
and a new aml_*() API that helps to create AML objects dynamically.

> 
>     Stefan
> 
>