Hi Ard,

On 9/14/26 1:06 PM, Ard Biesheuvel wrote:
> On Mon, 14 Sep 2026, at 02:57, Melody Wang wrote:
>> The SVSM APIC protocol supports 5 calls. SVSM_APIC_CONFIGURE_EMULATION
>> (shortened to SVSM_APIC_CONFIG_EMULATION for brevity), call 1, provides
>> the controls whether the guest can make use of the SVSM APIC protocol.
>>
>> Implement this call, and register Alternate Injection for the guest
>> by default.
>>
>> Signed-off-by: Melody Wang <huibo.wang@amd.com>
>> ---
>>   arch/x86/boot/compressed/sev.c | 17 +++++++++++++++++
>>   arch/x86/boot/compressed/sev.h |  6 ++++++
>>   2 files changed, 23 insertions(+)
>>
> 
> Does this need to happen in the EFI stub and/or decompressor? We have swathes
> of early SEV-SNP boot code in the kernel proper; could it be done there?

Let me answer the question directly first:

Alternate Injection is used by both the guest firmware and the guest 
kernel. Before we run the guest kernel, guest firmware - OVMF - is using 
Alternate Injection for interrupts, etc.

So ExitBootServices() is a handoff boundary. When it is called, the 
firmware's use of Alternate Injection ends. It must deregister its use 
of Alternate Injection at this point since it can not assume anything 
about the guest kernel that follows.

At this moment, if the guest kernel has not registered the use of 
Alternate Injection, latter is disabled and can not be re-enabled.

So if the guest kernel wants Alternate Injection, it has to happen 
before ExitBootServices().

I hope this answers the question, the rest below is a longer, more 
detailed explanation which you don't necessarily need to read - only if 
you're interested.

:)

Now here's the long explanation along with the spec on the matter:

There is an Alternate Injection spec (I'm attaching it because the 
mailing list archives which hosted it, are not up currently) and it says 
that Alternate Injection supports multiple guest runtimes, see the 
"Negotiating Functionality" section.

When there is a handoff from a UEFI virtual firmware to a guest 
operating system, the firmware, before terminating its execution must be 
prepared to disable the use of Alternate Injection unless it knows the 
guest kernel is prepared to continue use of Alternate Injection.

However, there is no mechanism for them to negotiate this directly, so 
the SVSM APIC protocol supports a registration mechanism to determine 
the behavior across handoff by using a registration counter.

The spec has an example how this mechanism might be used in the case of 
transition from UEFI-based firmware to a guest OS:

"Prior to executing ExitBootServices, a guest OS that understands the 
SVSM APIC Protocol should detect the availability of that protocol and 
determine whether it is usable. If so, it should register its use of the 
SVSM APIC Protocol.

During the execution of ExitBootServices, the firmware (which does not 
know the intention of the guest OS) should deregister its use of the 
SVSM APIC protocol. If the guest OS registered its intention to use 
Alternate Injection, this call will not affect the current state of 
Alternate Injection. If the guest OS did not declare its intention to 
use Alternate Injection, then this call will disable Alternate Injection."

During the handoff, the firmware needs to deresgister its use of 
Alternate Injection, so the registration counter can reflect the right 
status of using Alternate Injection for runtimes. Once the registration 
count reaches zero, Alternate Injection is disabled and cannot be 
re-enabled, so if the guest OS wants Alternate Injection, it has to 
register before ExitBootServices() to keep the count > 0.

I hope that makes sense.

-- 
Thanks,
Melody