On 06/03/2016 06:15 PM, Duc Dang wrote: > Do you have other suggestions? Otherwise, I will prepare a patch > following Lorenzo's approach. Duc has since left Applied for other pastures. I miss him, he's a great guy. He laid all the right groundwork for this, but the ACPI binding still needs to be upstreamed. It's a few lines of code matching on APMC0D0E but without it, upstream kernels wont have working MSI on X-Gene with ACPI. I need this to be upstreamed soon please :) Can someone at APM followup with an updated patch, and get it in? Here's the rub. The average person booting a Linux box (even a good kernel person) isn't going to say "hey, MSIs aren't setup right on this ARM server because it's compliant with 1 out of 3 possible ways MSIs might be done at a high level [let's forget the many others] and all it needs is this...". What they're going to say is "huh, PCIe card doesn't work, might be an MSI problem". Which is the email I have after someone tried using an IB card in an X-Gene box and spent a few hours poking. We're so close to having "ACPI all the things" but the latest development builds of RHEL don't do MSI on X-Gene because of the Red Hat "upstream first" rules. So let's get that fixed. Thanks, Jon. -- Computer Architect | Sent from my Fedora powered laptop
