[PATCH v2 01/12] ACPI/IORT: Make iort_match_node_callback walk the ACPI namespace for NC

[Lorenzo Pieralisi]

On Tue, Jun 30, 2020 at 11:06:41AM +0800, Hanjun Guo wrote:

[...]

> > For devices that aren't described in the DSDT - IORT translations
> > are determined by their ACPI parent device. Do you see/Have you
> > found any issue with this approach ?
> 
> The spec says "Describes the IO relationships between devices
> represented in the ACPI namespace.", and in section 3.1.1.3 Named
> component node, it says:

PCI devices aren't necessarily described in the ACPI namespace and we
still use IORT to describe them - through the RC node.

> "Named component nodes are used to describe devices that are also
> included in the Differentiated System Description Table (DSDT). See
> [ACPI]."
> 
> So from my understanding, the IORT spec for now, can only do ID
> translations for devices in the DSDT.

I think you can read this multiple ways but this patch does not
change this concept. What changes, is applying parent's node IORT
mapping to child nodes with no associated DSDT nodes, it is the
same thing we do with PCI and the _DMA method - we could update
the wording in the specs if that clarifies but I don't think this
deliberately disregards the specifications.

> > > For a platform device, if I use its parent's full path name for
> > > its named component entry, then it will match, but this will violate
> > > the IORT spec.
> > 
> > Can you elaborate on this please I don't get the point you
> > are making.
> 
> For example, device A is not described in DSDT so can't represent
> as a NC node in IORT. Device B can be described in DSDT and it
> is the parent of device A, so device B can be represented in IORT
> with memory access properties and node flags with Substream width
> and Stall supported info.
> 
> When we trying to translate device A's ID, we reuse all the memory
> access properties and node flags from its parent (device B), but
> will it the same?

I assume so why wouldn't it be ? Why would be describe them in
a parent-child relationship if that's not how the system looks like
in HW ?

Do you have a specific example in mind that we should be aware of ?

> So the IORT spec don't support this, at least it's pretty vague
> I think.

I think that's a matter of wording, it can be updated if it needs be,
reach out if you see any issue with the current approach please.

Thanks,
Lorenzo