Re: [edk2-devel] [edk2-platforms] [PATCH v1 2/3] Platform/ARM/SgiPkg: Add HMAT ACPI table for RdN1EdgeX2

["Jonathan Cameron" <jonathan.cameron@huawei.com>]

On Fri, 29 Jan 2021 17:11:03 +0000
Vijayenthiran Subramanian <vijayenthiran.subramaniam@arm.com> wrote:

Hi Vijayenthiran,

....

> > > > +  // Memory Proximity Domain
> > > > +  {
> > > > +    EFI_ACPI_6_3_HMAT_STRUCTURE_MEMORY_PROXIMITY_DOMAIN_ATTRIBUTES_INIT (
> > > > +      1, 0x0, 0x0),
> > > > +    EFI_ACPI_6_3_HMAT_STRUCTURE_MEMORY_PROXIMITY_DOMAIN_ATTRIBUTES_INIT (
> > > > +      1, 0x1, 0x1),
> > > > +   },
> > > > +
> > > > +  // Latency Info
> > > > +  EFI_ACPI_6_3_HMAT_STRUCTURE_SYSTEM_LOCALITY_LATENCY_AND_BANDWIDTH_INFO_INIT (
> > > > +    0, 0, INITATOR_PROXIMITY_DOMAIN_CNT, TARGET_PROXIMITY_DOMIAIN_CNT, 100),
> > > > +  {
> > > > +    {0, 1}, {0, 1},
> > > > +    {
> > > > +      10, 20,
> > > > +      20, 10,
> > > > +    }
> > > > +  },  
> >
> > Hi,
> >
> > Now either you have a very odd machine in which latencies are round numbers that happen
> > to look like the values in SLIT or ... ?
> >
> > HMAT has very carefully defined real world units to overcome the fact that there
> > was no standard scaling etc for SLIT.  It is not a good idea to put tables out
> > there which don't do this right as someone may copy this for a real system
> > and we end up with HMAT being roughly as useless for performance prediction as
> > SLIT is.  That is not something I would want to see.
> >
> > Jonathan  
> 
> This patch is for a platform that is modelled and not for an actual board. So
> the latencies are hypothetical but ensure that the typical latency between two
> chips is represented. These tables are not expected to be copied for another
> platform without understanding what these values mean. Such a copy, without
> understanding what is being copied, would be an error in judgement on the part
> of the person doing it. The commit message of this patch (in v2) can be updated
> to exercise caution on reusing these values.
> 
> [1]: https://developer.arm.com/products/system-design/fixed-virtual-platforms

No one reads commit messages.  So if you want to avoid people copying them put a
comment in the actual file.  Or just make up some more plausible numbers :)
All we really need here is to avoid giving the impression that copying numbers from
SLIT is valid.

Jonathan


> 
> Best Regards,
> Vijayenthiran
> 
> >  
> > > > +
> > > > +  // Memory Side Cache
> > > > +  EFI_ACPI_6_3_HMAT_STRUCTURE_MEMORY_SIDE_CACHE_INFO_INIT (
> > > > +    0x0, SIZE_8MB,
> > > > +    HMAT_STRUCTURE_MEMORY_SIDE_CACHE_INFO_CACHE_ATTRIBUTES_INIT (
> > > > +      1,
> > > > +      1,
> > > > +      2,
> > > > +      2,
> > > > +      64 // 64 bytes cache line length
> > > > +    ),
> > > > +    0),
> > > > +
> > > > +  EFI_ACPI_6_3_HMAT_STRUCTURE_MEMORY_SIDE_CACHE_INFO_INIT (
> > > > +    0x1, SIZE_8MB,
> > > > +    HMAT_STRUCTURE_MEMORY_SIDE_CACHE_INFO_CACHE_ATTRIBUTES_INIT (
> > > > +      1,
> > > > +      1,
> > > > +      2,
> > > > +      2,
> > > > +      64 // 64 bytes cache line length
> > > > +    ),
> > > > +    0),
> > > > +};
> > > > +
> > > > +VOID* CONST ReferenceAcpiTable = &Hmat;  
> > >
> > >
> > >
> > >
> > >
> > >