Re: [RFC 00/14] AMD: Add Secure AVIC Guest Support

From: Neeraj Upadhyay
Date: Tue Oct 29 2024 - 08:15:54 EST




On 10/29/2024 5:21 PM, Kirill A. Shutemov wrote:
> On Tue, Oct 29, 2024 at 03:54:24PM +0530, Neeraj Upadhyay wrote:
>> diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c
>> index aeda74bf15e6..08156ac4ec6c 100644
>> --- a/arch/x86/kernel/apic/apic.c
>> +++ b/arch/x86/kernel/apic/apic.c
>> @@ -1163,6 +1163,9 @@ void disable_local_APIC(void)
>> if (!apic_accessible())
>> return;
>>
>> + if (apic->teardown)
>> + apic->teardown();
>> +
>> apic_soft_disable();
>>
>> #ifdef CONFIG_X86_32
>
> Hm. I think it will call apic->teardown() for all but the one CPU that
> does kexec. I believe we need to disable SAVIC for all CPUs.
>

I see it being called for all CPUs.

For the CPU doing kexec, I see below backtrace, which lands into disable_local_APIC()

disable_local_APIC
native_stop_other_cpus
native_machine_shutdown
machine_shutdown
kernel_kexec

For the other CPUs, it is below:

disable_local_APIC
stop_this_cpu
__sysvec_reboot
sysvec_reboot


> Have you tested the case when the target kernel doesn't support SAVIC and
> tries to use a new interrupt vector on the boot CPU? I think it will
> break.
>

For a VM launched with VMSA feature containing Secure AVIC, the target
kernel also is required to support Secure AVIC. Otherwise, guest bootup
would fail. I will capture this information in the documentation.
So, as far as I understand, SAVIC kernel kexecing into a non-SAVIC kernel
is not a valid use case.



- Neeraj