Add a generic ahci attachment

[andrew]

Summary

This adds support for AHCI to attach to a generic bus. On arm64 this is
on the internal memory bus and will appear as a memory-mapped device.

While this is currently FDT specific I expect this to be reused with ACPI
as this is an option on ARMv8 servers.

[mav]

I have no real objections, other then it looks suspiciously simplistic: no quirks, no suspend-resume, no setting for some controller properties (I am not sure newbus will always zero softc), ...

sys/dev/ahci/ahci_generic.c

57 ↗

(On Diff #10473)

Why is this function needed at all?

[andrew]

In D4269#89984, @mav wrote:

I have no real objections, other then it looks suspiciously simplistic: no quirks, no suspend-resume, no setting for some controller properties (I am not sure newbus will always zero softc), ...

There are no quirks because we won't hit the only know errata on the hardware I tested this on. It's related to the second sata port which is disabled meaning the driver never sees it.

I left out suspend/resume as I it's untested, none of the suspend/resume paths have been tested as there is currently nothing similar to the Cx idle states on arm64.

The device_get_softc man page datest "The softc is automatically allocated and zeroed when the device is attached".

sys/dev/ahci/ahci_generic.c

57 ↗

(On Diff #10473)

The larger patch I pulled this from did some extra work here that seems to not be needed on the hardware I have.

[andrew]

Remove an unneeded call to ofw_bus_get_node

[mmel]

Please, see inline comment. The rest is OK for me.

sys/conf/files.arm64

60 ↗

(On Diff #10478)

Use other name that ahci_fdt. ahci_generic (although is still ugly) at least matched with driver name.

[mmel]

sys/conf/files.arm64

60 ↗

(On Diff #10478)

Taking back, i misread option name as "ahci_fdt" (with underscore). Sorry.