Acpi Rewrite - AML: Problem resolving PCI Routing Table entries

[newcomb-luke]

I am making this issue for the new ACPI crate rewrite as it seems like that is where things will be going.

Using QEMU / OVMF it is impossible to use the current AML implementation using the 'pc' virtual machine.

When a PCI device wants to specify a link device, when running in 'pc' mode the resolved type is:

Reference {
    kind: RefOf,
    inner: WrappedObject(Device)
}

The old implementation would resolve this to a device name string which could then be used to look for the _CRS entry on the link device. However, now because it resolves to a Device, a dead end is reached.

When using the 'q35' device, this problem does go away, as that device seems to be what is used to test this implementation, and the reference kind turns into Unresolved which then resolves to a String.

It is important to note that in the current (non-rewritten) ACPI crate, this works as expected. This is just present in the new version.

[IsaacWoods]

Hi, thanks for reporting this! Looks like the old tables differ from q35's in that they define the link objects before the _PRT, which causes the object to actually be resolved, which makes it impossible to find the path, which obviously needs to be addressed.

The NT interpreter seems to simply not resolve paths at all within packages, instead passing them out as strings. They can then be resolved (wherever they are defined) at time of use of the package. I didn't like this approach at first (it does still feel a bit hacky), but it's probably worth just emulating NT here to maximise table compatibility, strings and all.

I'm currently abroad with only a laptop, but will see if I have a chance to play around with this (will also need to rewrite the PRT parsing code), or you're very welcome to open a PR with this change yourself.

It is important to note that in the current (non-rewritten) ACPI crate, this works as expected. This is just present in the new version.

The old interpreter unfortunately is of a poor design and will not be maintained. While the new version may still have issues with AML the old one accepted, the new version is very much more maintainable and is the right way forward, and we'll need to correct its behaviour in those few cases. The PR has spiralled in scope to a full reorganisation of the crates as well as a rewrite of large parts of both crates, which I unfortunately didn't get merged before travelling for a few months. Apologies for confusion around which version to use at this time.

Thanks for trying out the new version and please continue to report issues you find.