Re: [patch V2 50/58] x86/apic: Provide common init infrastructure

From: Thomas Gleixner
Date: Mon Jul 31 2023 - 09:01:31 EST


On Mon, Jul 31 2023 at 13:31, Juergen Gross wrote:
> On 24.07.23 15:35, Thomas Gleixner wrote:
>> static void __init xen_apic_check(void)
>> {
>> - if (apic == &xen_pv_apic)
>> - return;
>> -
>> - pr_info("Switched APIC routing from %s to %s.\n", apic->name,
>> - xen_pv_apic.name);
>> - apic = &xen_pv_apic;
>> + apic_install_driver(&xen_pv_apic);
>> }
>> +
>> void __init xen_init_apic(void)
>> {
>> x86_apic_ops.io_apic_read = xen_io_apic_read;
>> /* On PV guests the APIC CPUID bit is disabled so none of the
>> * routines end up executing. */
>> if (!xen_initial_domain())
>> - apic = &xen_pv_apic;
>> + apic_install_driver(&xen_pv_apic);
>
> This is working, but it produces a WARN() splat when booting as an unprivileged
> Xen PV guest from static_call patching (static_call_init() hasn't been called
> yet).

Duh, yes. It's too early.

> The diff below on top is fixing the issue:
>
> diff --git a/arch/x86/xen/apic.c b/arch/x86/xen/apic.c
> index 1838aefc632f..84f24268670b 100644
> --- a/arch/x86/xen/apic.c
> +++ b/arch/x86/xen/apic.c
> @@ -163,14 +163,18 @@ static void __init xen_apic_check(void)
> apic_install_driver(&xen_pv_apic);
> }
>
> +void __init xen_apic_install(void)
> +{
> + /*
> + * On PV guests the APIC CPUID bit is disabled so none of the
> + * routines end up executing.
> + */
> + apic_install_driver(&xen_pv_apic);
> +}
> +
> void __init xen_init_apic(void)
> {
> x86_apic_ops.io_apic_read = xen_io_apic_read;
> - /* On PV guests the APIC CPUID bit is disabled so none of the
> - * routines end up executing. */
> - if (!xen_initial_domain())
> - apic_install_driver(&xen_pv_apic);
> -
> x86_platform.apic_post_init = xen_apic_check;
> }
> apic_driver(xen_pv_apic);

I wonder whether this explicit install is actually needed at all.
Shouldn't the driver be installed via the APIC probing mechanism
automagically?

Thanks,

tglx