On Wednesday 05 November 2014 20:32:19 al.stone@linaro.org wrote:
From: Al Stone ahs3@redhat.com
This patch set is in the acpi-topic-seattle branch of acpi.git.
The basic idea is that these patches are an example of how to implement ACPI for the AMD Seattle platform. Many of them have been borrowed from the Fedora arm64 kernel tree which is at:
git://git.fedorahosted.org/git/kernel-arm64.git
Additional patches come from Linaro and LKMLA. With the exception of the PCI patches, or others mentioned below, the intent is that the original authors of these patches will submit them upstream. As always, they rely on the ACPI core patches already being in place.
Looks good for the most part, thanks a lot for posting the series!
The majority of the patches are not actually Seattle specific as far as I can tell, so I think it would be good to separate them into another series, or put them into the base support series.
The most controversial parts that I can see are the X-Gene specific patches, parking protocol and PCI hacks in particular, which IMO have no business in this series.
Patch 0026 adds ACPI support to the AMD XGBE 10Gbe interface. Yes, just one patch.
This is probably the only Seattle specific patch that we need to discuss in more detail. I've replied with some comments on the issues I see here, but I think we should take that discussion to a broader mailing list.
These patches were tested on an AMD Seattle platform, using the 71C version of their firmware (future versions will have needed fixes for PCI windows and some other odds and ends).
Could you also include an email with the DSDT source from that version? There might be some more things in there that we need to discuss that were not obvious from the series.
Things to do: -- Get PCI re-done; there are way too many x86-isms in the code, and some serious refactoring and cleanup is needed; the ia64 seems to have taken some good steps in this direction.
From the discussions we had on the ARM64 PCI support, I think it would
be better to leave x86 and ia64 alone for now and put the new generic code into drivers/pci/. x86 can migrate to that over time if they want to, for ia64 there is probably little point in touching it. ia64 also handles things like I/O space in its own way that is incompatible with most other architectures.
-- MSI(-X) support: a DT version has been posted, and some ACPI work is under way.
I think we are still lacking a binding for GICv3+PCI here even in the DT case. We should definitely come up with a way to describe the translation that works on both DT and ACPI. The problem is that the mapping of domain/bus/device/function numbers to GIC requester IDs is not standardized in hardware, so anyone can do it differently and we need to describe the mapping.
-- KVM support: ACPI support patches are available, but were not included here since I have not had time to test them.
Is this for host or guest?
Arnd