Re: [PATCH v3 07/10] hw/i386: declare ACPI mother board resource for MMCONFIG region

[Isaku Yamahata]

On Tue, Feb 16, 2021 at 08:45:48AM -0500,
"Michael S. Tsirkin" <mst@redhat.com> wrote:

> On Tue, Feb 16, 2021 at 01:43:01AM -0800, Isaku Yamahata wrote:
> > On Mon, Feb 15, 2021 at 01:48:32PM +0100,
> > Igor Mammedov <imammedo@redhat.com> wrote:
> > 
> > > On Fri, 12 Feb 2021 12:51:57 -0800
> > > Isaku Yamahata <isaku.yamahata@gmail.com> wrote:
> > > 
> > > > On Fri, Feb 12, 2021 at 04:40:38PM +0100,
> > > > Igor Mammedov <imammedo@redhat.com> wrote:
> > > > 
> > > > > On Wed, 10 Feb 2021 22:46:43 -0800
> > > > > Isaku Yamahata <isaku.yamahata@intel.com> wrote:
> > > > >   
> > > > > > +    Aml *dev;
> > > > > > +    Aml *rbuf;
> > > > > > +    Aml *resource_template;
> > > > > > +    Aml *rbuf_name;
> > > > > > +    Aml *crs;
> > > > > > +
> > > > > > +    if (!acpi_get_mcfg(&mcfg)) {
> > > > > > +        return NULL;
> > > > > > +    }
> > > > > > +
> > > > > > +    /* DRAM controller */
> > > > > > +    dev = aml_device("DRAC");
> > > > > > +    aml_append(dev, aml_name_decl("_HID", aml_string("PNP0C01")));
> > > > > > +
> > > > > > +    resource_template = aml_resource_template();
> > > > > > +    aml_append(resource_template,
> > > > > > +               aml_qword_memory(AML_POS_DECODE,
> > > > > > +                                AML_MIN_FIXED,
> > > > > > +                                AML_MAX_FIXED,
> > > > > > +                                AML_NON_CACHEABLE,
> > > > > > +                                AML_READ_WRITE,
> > > > > > +                                0x0000000000000000,
> > > > > > +                                mcfg.base,
> > > > > > +                                mcfg.base + mcfg.size - 1,  
> > > > > s/mcfg.base + mcfg.size - 1/mcfg.base/  
> > > > 
> > > > AddressMaximum needs to be the highest address of the region.
> > > > Not base address. By disassemble/assembl it, iasl complains as follows.
> > > > Also there are similar examples in acpi-build.c.
> > > I mostly clean up all places to use the same base address for min/max,
> > > but probably something were left behind.
> > > 
> > > spec says:
> > > 
> > > acpi 6.3: 19.6.110 QWordMemory
> > > 
> > > AddressMaximum evaluates to a 64-bit integer that specifies the highest 
> > > possible base address of the
> > > Memory range. The value must have ‘0’ in all bits where the corresponding 
> > > bit in AddressGranularity is
> > > ‘1’. For bridge devices which translate addresses, this is the address on 
> > > the secondary bus. The 64-bit
> > > field DescriptorName ._MAX is automatically created to refer to this 
> > > portion of the resource descriptor.
> > 
> > Ok, Linux guest is happy with min=max.
> > I conlude that it's iasl issue.
> > 
> > Thanks,
> 
> OK but what about all the other places in the code that seem to use this
> field differently?

Igor, what do you think?
The followings are the summary of the situation.

- acpi 6.4: 19.6.110 QWordMemory
  _MAX: maximum of base address.

- acpi 6.4: 6.4.3.5 Address Space Resource Descriptors
  table 6.44
  If _LEN > 0 and _MIF = 1 and _MAF = 1, then _LEN = _MAX - _MIN + 1
  This doesn't match with the above description

- iasl
  If _LEN > 0 and _MIF = 1 and _MAF = 1,
  it emits warning on _LEN != _MAX - _MIN + 1

- Linux Guest MMCONFIG check 
  check_mcfg_resoure() uses only _MIN. doesn't use _MAX.
  _MAX value doesn't matter

- Linux acpi code
  acpi_decode_space() uses _MAX to calulate range [start, end], not use _LEN.
  i.e. It assumes _LEN = _MAX - _MIN + 1 if _LEN > 0, _MIF = 1, _MAF = 1

- qemu code sets for [qd]word_memory_resource (except this patch)
  If _LEN > 0 and _MIF = 1 and _MAF = 1, then set _LEN = _MAX - _MIN + 1

-- 
Isaku Yamahata <isaku.yamahata@gmail.com>