Re: [PATCH 2/7] x86/apic: Add an SVSM APIC driver

[Tom Lendacky]

On 8/20/26 00:06, Melody Wang wrote:
> Hi Tom,

Hi Melody,

> 
> On 7/31/26 10:39 AM, Tom Lendacky wrote:
>>> +int svsm_do_call(struct svsm_call *call)
>>> +{
>>> +    call->caa = svsm_get_caa();
>>> +    return svsm_perform_call_protocol(call);
>>> +}
>>
>> This helper is unneeded, all it does is set the calling area and then
>> issue svsm_perform_call_protocol(). You're already setting other call
>> values where you use this helper, so just add setting the calling area to
>> those and directly issue svsm_perform_call_protocol().
> 
> Since svsm_perform_call_protocol() and svsm_get_caa() are all internal
> functions to coco (arch/x86/coco/sev/internal.h), I can not call them
> directly from the apic driver, so perhaps it is better that I have this
> helper function here.

Why not move svsm_get_caa() and svsm_perform_call_protocol() from
internal.h then and make them available? Or create a callable function
that lives in arch/x86/coco/sev/svsm.c that builds the svsm_call struct
and performs the SVSM call?
> 
>>> diff --git a/arch/x86/kernel/apic/svsm_apic.c b/arch/x86/kernel/apic/
>>> svsm_apic.c
>>> new file mode 100644
>>> +static int svsm_apic_probe(void)
>>> +{
>>> +    if (!cc_platform_has(CC_ATTR_SNP_ALTERNATE_INJECTION) || !snp_vmpl)
>>
>> Do you need both checks? Shouldn't the attribute only be set if the
>> feature is set and the VMPL level is greater than 0?
> 
> Yes, the cc_platform_has() just checks if the sev_status has the
> MSR_AMD64_SNP_ALTERNATE_INJ bit set.

Isn't that all that is needed? If the attribute is set none of the other
injection methods can be used. If anything, you should terminate if
alternate injection is enabled and you are running at VMPL0, because
nothing can update the VMSA to set the injection/irq fields.

> 
>>> +    /* Alternate Injection and Secure AVIC are mutually exclusive */
>>> +    if (cc_platform_has(CC_ATTR_SNP_SECURE_AVIC))
>>> +        return 0;
>>
>> What happens in this situation? Will Secure AVIC still work if Alternate
>> Injection is enabled?
> 
> Secure AVIC and Alternate Injection are mutually exclusive, if the
> sev_status has secure AVIC set, we should not probe svsm apic:
> 
> "15.36.21.2 VMRUN and #VMEXIT
> Secure AVIC mode is mutually exclusive with Restricted Injection,
> Alternate Injection, and
> hypervisor controlled AVIC modes. If the SecureAvic bit is set to 1, and
> the AVIC Enable bit in the
> VMCB is set to 1 or the RestrictedInjection or AlternateInjection bits
> in SEV_FEATURES are set to 1,
> VMRUN will fail with #VMEXIT(VMEXIT_INVALID)."
> 
> I'm thinking I should probably check for secure AVIC first and if set,
> return 0.

If the VMRUN fails because both are set, how can you possibly be running
in the guest with both set? So I see no need to check for Secure AVIC.
> 
>>> +
>>> +    if (!x2apic_mode) {
>>> +        pr_err("Alternate Injection in non x2APIC mode impossible.
>>> Terminating.\n");
>>> +        sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED);
>>> +    }
>>> +
>>> +    pr_info("Alternate Injection SVSM APIC enabled\n");
>>> +
>>
>> Should the probe routine query the SVSM for the APIC protocol?
> 
> Right now, the query for the APIC protocol is hardcoded as 0 which means
> basic APIC functionality related to interrupt delivery. So it is not

What if the version of the SVSM that is running doesn't have the APIC
emulation protocol?

> needed to query the APIC protocol now. In the future, when the SVSM code
> changes with different set, we can adjust the guest code accordingly.
> 
>>
>>> +    return 1;
>>> +}
>>> +
>>> +static int svsm_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
>>> +{
>>> +    return x2apic_enabled() &&
>>> cc_platform_has(CC_ATTR_SNP_ALTERNATE_INJECTION) && snp_vmpl;
>>> +}
>>> +
>>> +static void svsm_apic_msr_write(u32 reg, u32 v)
>>> +{
>>> +    u32 msr = APIC_BASE_MSR + (reg >> 4);
>>> +    struct svsm_call call = {};
>>> +    int ret;
>>> +
>>> +    switch (reg) {
>>> +    case APIC_ID:
>>> +    case APIC_TASKPRI:
>>> +    case APIC_PROCPRI:
>>> +    case APIC_EOI:
>>> +    case APIC_ISR ... APIC_ISR + 0x70:
>>> +    case APIC_TMR ... APIC_TMR + 0x70:
>>> +    case APIC_IRR ... APIC_IRR + 0x70:
>>> +    case APIC_ICR:
>>> +    case APIC_SELF_IPI:
>>> +        call.rax = SVSM_APIC_CALL(SVSM_APIC_WRITE_REGISTER);
>>> +        call.rcx = msr;
>>> +        call.rdx = v;
>>> +
>>> +        ret = svsm_do_call(&call);
>>> +        if (ret) {
>>> +            pr_err("SVSM_APIC_WRITE_REGISTER: 0x%x, error: %d\n",
>>> reg, ret);
>>> +            sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED);
>>
>> Can this be a new SEV_TERM_SET_LINUX value?
>>
>>> +        }
>>> +        break;
>>> +    default:
>>> +        pr_err("SVSM_APIC_WRITE_REGISTER 0x%x not supported\n", reg);
>>> +        break;
>>> +    }
>>> +}
>>> +
>>> +static u32 svsm_apic_msr_read(u32 reg)
>>> +{
>>> +    u32 msr = APIC_BASE_MSR + (reg >> 4);
>>> +    struct svsm_call call = {};
>>> +    int ret;
>>> +
>>> +    switch (reg) {
>>> +    case APIC_ID:
>>> +    case APIC_TASKPRI:
>>> +    case APIC_PROCPRI:
>>> +    case APIC_EOI:
>>> +    case APIC_ISR ... APIC_ISR + 0x70:
>>> +    case APIC_TMR ... APIC_TMR + 0x70:
>>> +    case APIC_IRR ... APIC_IRR + 0x70:
>>> +    case APIC_ICR:
>>> +    case APIC_SELF_IPI:
>>> +        call.rax = SVSM_APIC_CALL(SVSM_APIC_READ_REGISTER);
>>> +        call.rcx = msr;
>>> +
>>> +        ret = svsm_do_call(&call);
>>> +        if (ret) {
>>> +            pr_err("SVSM_APIC_READ_REGISTER: 0x%x, error: %d\n",
>>> reg, ret);
>>> +            sev_es_terminate(SEV_TERM_SET_GEN, GHCB_SNP_UNSUPPORTED);
>>> +        }
>>> +        break;
>>> +    default:
>>> +        pr_err("SVSM_APIC_READ_REGISTER: 0x%x not supported\n", reg);
>>> +        return 0;
>>> +    }
>>> +
>>> +    return call.rdx_out;
>>> +}
>>
>> The read and write are very similar. Can you have a common function that
>> takes a reg paramter, value parameter (that is input and output), and a
>> mode parameter (read/write) and then have small read and write functions?
> 
> Yes, and Sashiko pointed that I need to prevent preemption for the caa
> call, I agree with it, but I think I should prevent interrupts here - I
> should do native_local_irq_save(), because there should not be any
> interrupts during a caa call as those things are not reentrant. Thoughts?

svsm_perform_call_protocol() already disables interrupts.

Thanks,
Tom

> 
> Thanks,
> Melody