On 20/06/14 16:57, Joel Schopp wrote:
>
> On 06/20/2014 09:23 AM, Andre Przywara wrote:
>>
>> On 20/06/14 15:14, Joel Schopp wrote:
>>> On a related note I've just joined AMD and am working on enabling kvm on
>>> an aarch64 SOC with a gic-400. Looking at upstream it appears that only
>>> the cortex-a15-gic has been merged before on arm. I expect enabling the
>> cortex-a15-gic is actually kind of a misnomer for a generic GICv2. So
>> any generic GICv2 (or GIC-400) should be named like this and it should
>> work. It definitely does on the ARM hardware I have here.
> It's my understanding that the gic-400 is GICv2 for the virtualization
> bits but has some other extensions to GICv2 in other areas such that
> will be necessary to represent it in the device tree as a new gic. I
> was able to boot with kvm enabled (but not actually working properly) by
> just doing the following:
>
> In the dts:
> gic: interrupt-controller@e1101000 {
> compatible = "arm,gic-400";
> #interrupt-cells = <3>;
> #address-cells = <0>;
> interrupt-controller;
> msi-controller;
> reg = <0x0 0xe1110000 0 0x1000>, /* gic dist */
> <0x0 0xe112f000 0 0x1000>, /* gic cpu */
> <0x0 0xe1180000 0 0x1000>, /* gic msi */
Bang. This is not compliant with the normal binding. I suppose what you
have is GICv2m (basically GICv2 + MSI support extension). You shouldn't
describe this as part of the GIC itself, but as a separate node.
> <0x0 0xe1140000 0 0x2000>, /* gic virtual ic*/
> <0x0 0xe116f000 0 0x2000>; /* gic virtual cpu*/
>
> interrupts = <1 8 0xf04>;
> };
>
>
> --- a/virt/kvm/arm/vgic.c
> +++ b/virt/kvm/arm/vgic.c
> @@ -1478,9 +1478,13 @@ int kvm_vgic_hyp_init(void)
>
> vgic_node = of_find_compatible_node(NULL, NULL,
> "arm,cortex-a15-gic");
> if (!vgic_node) {
> - kvm_err("error: no compatible vgic node in DT\n");
> - return -ENODEV;
> + vgic_node = of_find_compatible_node(NULL, NULL,
> "arm,gic-400");
> }
> + if (!vgic_node) {
> + kvm_err("error: no compatible vgic node in DT\n");
> + return -ENODEV;
> + }
> +
If you want to match multiple compat strings, you should consider using
of_find_matching_node_and_match.
>>
>>> gic400 virtualization bits will be a small mount of code compared to
>>> your GICv3 changes, but thought I'd chime in as I'd be happy to be ccd
>>> on and review any gic related patches.
>> Is there any specific issue you are encountering?
>>
> I'm really just starting at this and don't feel like I've done enough
> investigation to even know what the problem is. That said here's how
> the issue I'm seeing manifests itself:
>
> When I try to run qemu it outputs:
> error: kvm run failed Bad address Aborted
>
> KVM throws this into /var/log/messages (0x20 is HSR_EC_IABT):
> Jun 19 14:05:08 joelaarch64 kernel: kvm [1553]: Unsupported fault status: EC=0x20 DFCS=0x14
> Jun 19 14:05:08 joelaarch64 kernel: kvm [1553]: Unsupported fault status: EC=0x20 DFCS=0x14
>
> It's probably just a stupid mistake on my part like setting up the
> device tree incorrectly. It will take me a bit more digging to even get
> to the root cause .
>
Yeah, all kind of ugly things will happen if you mess with the binding.
Just remove this extra line in your DT and it should start working
nicely (provided that you fix your compat string as well).
M.
--
Jazz is not dead. It just smells funny...
_______________________________________________
kvmarm mailing list
kvmarm@xxxxxxxxxxxxxxxxxxxxx
https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
