Re: [Xen-devel] [PATCH] xen/acpi-processor: C and P-state driver that uploads said data to hypervisor.

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 13.03.12 at 04:35, Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx> wrote:
> +static unsigned int __init get_max_acpi_id(void)
> +{
> +     struct xenpf_pcpuinfo *info;
> +     struct xen_platform_op op = {
> +             .cmd = XENPF_get_cpuinfo,
> +             .interface_version = XENPF_INTERFACE_VERSION,
> +     };
> +     int ret = 0;
> +     unsigned int i, max_apic_id, max_acpi_id = 0;
> +
> +     info = &op.u.pcpu_info;
> +     info->xen_cpuid = 0;
> +
> +     ret = HYPERVISOR_dom0_op(&op);
> +     if (ret)
> +             return max_acpi_id;
> +
> +     /* The max_present is the same irregardless of the xen_cpuid */
> +     max_apic_id = op.u.pcpu_info.max_present;

The 'apic' in the name here is confusing, as this isn't an APIC ID, but
(Xen) ID of the last present CPU (i.e. one less than the maximum
count).

> +     for (i = 0; i <= max_apic_id; i++) {
> +             info->xen_cpuid = i;
> +             ret = HYPERVISOR_dom0_op(&op);
> +             if (ret)
> +                     continue;
> +             max_acpi_id = max(info->acpi_id, max_acpi_id);
> +     }
> +     pr_debug(DRV_NAME "Max ACPI ID: %u\n", max_acpi_id);
> +     return max_acpi_id;

Iirc hotplug CPUs may not be represented in MADT, and hence I'd
suggest adding some slack here (e.g. doubling the value).

> +}

Now, if the whole thing could become a true extension to processor.ko,
the fishy (and too weak in the 'm' case, as you noted yourself) Kconfig
dependency on the cpufreq drivers could go away, since both
acpi-cpufreq and powernow-k8 depend on ACPI_PROCESSOR. Which
would at once put it on par with the XenoLinux placement (in terms of
the point in time when it gets executed) of the code.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel