On 11/11/2014 07:27 AM, Arnd Bergmann wrote:
On Monday 10 November 2014 19:46:47 al.stone@linaro.org wrote:
The following patches are pretty much the same core as used for the Seattle topic branch. There are a few additional APM X-Gene specific patches.
These are not final by any stretch of the imagination. As with the Seattle patches, there is much work to be done for PCI (especially since the Mustang hardware is a very special case), and it is a known the the network card does not yet work properly.
So why commit these patches? Simply put, to record the current progress and make visible the current work on Mustang.
This kernel does boot and run, and PCI does seem to work. As noted, the NIC is not yet correct but this is being investigated and may be firmware related, not kernel.
Thanks for posting these. I think it is very useful to see the specific problems you run into when trying to run ACPI on some hardware that isn't designed against the SBSA or with ACPI in mind, and what kinds of workarounds would be necessary if we wanted to support it.
I've replied to the parts that I see as most problematic. My impression is that overall, it's too much of a hack, and we are better off by never merging any of the X-Gene specific patches from your series upstream.
There is a theoretical way to support this, depending on what your reasons are for doing this work: the firmware could implement an EL2 hypervisor that implements an SBSA compliant guest by faking a PCI mmconfig register set and a standard AHCI and UART, plus PSCI mode. The obvious downside of this would be the inability to run virtual machines.
Arnd
Hey, Arnd. I'm still sorting comments, but I have had several other things take precedence. So, my apologies for not having responded already. I will as soon as I can, though.
And yeah, I'm not convinced this should ever go upstream, predominantly because of PCI. My plan is to do one more pass of this as a topic in the Linaro acpi.git tree and then I'll likely let it rest so I can stay focused on servers.