The ccp driver can be overloaded even with guest request rate limits.
The return value of -EBUSY means that there is no firmware error to
report back to user space, so the guest VM would see this as
exitinfo2 = 0. The false success can trick the guest to update its
message sequence number when it shouldn't have.

Instead, when ccp returns -EBUSY, that is reported to userspace as the
throttling return value.

Cc: Thomas Lendacky <Thomas.Lendacky@amd.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Joerg Roedel <jroedel@suse.de>
Cc: Peter Gonda <pgonda@google.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Sean Christopherson <seanjc@google.com>

Signed-off-by: Dionna Glaze <dionnaglaze@google.com>
---
 arch/x86/kvm/svm/sev.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
index e45f0cfae2bd..0ceb7e83a98d 100644
--- a/arch/x86/kvm/svm/sev.c
+++ b/arch/x86/kvm/svm/sev.c
@@ -4060,6 +4060,11 @@ static int snp_handle_guest_req(struct vcpu_svm *svm, gpa_t req_gpa, gpa_t resp_
 	 * the PSP is dead and commands are timing out.
 	 */
 	ret = sev_issue_cmd(kvm, SEV_CMD_SNP_GUEST_REQUEST, &data, &fw_err);
+	if (ret == -EBUSY) {
+		svm_vmgexit_no_action(svm, SNP_GUEST_ERR(SNP_GUEST_VMM_ERR_BUSY, fw_err));
+		ret = 1;
+		goto out_unlock;
+	}
 	if (ret && !fw_err)
 		goto out_unlock;
 
-- 
2.50.0.rc0.642.g800a2b2222-goog

On 6/5/25 10:02, Dionna Glaze wrote:
> The ccp driver can be overloaded even with guest request rate limits.
> The return value of -EBUSY means that there is no firmware error to
> report back to user space, so the guest VM would see this as
> exitinfo2 = 0. The false success can trick the guest to update its
> message sequence number when it shouldn't have.

-EBUSY from the CCP driver is an error, not a throttling condition. Either
the driver has marked the ASP/PSP as dead or there are no command buffers
available, which is an error situation. There is no throttling support in
the CCP driver. A mutex is used to serialize requests, but all requests
proceed at some point. So there should not be a special check for -EBUSY.

Thanks,
Tom

> 
> Instead, when ccp returns -EBUSY, that is reported to userspace as the
> throttling return value.
> 
> Cc: Thomas Lendacky <Thomas.Lendacky@amd.com>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: Joerg Roedel <jroedel@suse.de>
> Cc: Peter Gonda <pgonda@google.com>
> Cc: Borislav Petkov <bp@alien8.de>
> Cc: Sean Christopherson <seanjc@google.com>
> 
> Signed-off-by: Dionna Glaze <dionnaglaze@google.com>
> ---
>  arch/x86/kvm/svm/sev.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> index e45f0cfae2bd..0ceb7e83a98d 100644
> --- a/arch/x86/kvm/svm/sev.c
> +++ b/arch/x86/kvm/svm/sev.c
> @@ -4060,6 +4060,11 @@ static int snp_handle_guest_req(struct vcpu_svm *svm, gpa_t req_gpa, gpa_t resp_
>  	 * the PSP is dead and commands are timing out.
>  	 */
>  	ret = sev_issue_cmd(kvm, SEV_CMD_SNP_GUEST_REQUEST, &data, &fw_err);
> +	if (ret == -EBUSY) {
> +		svm_vmgexit_no_action(svm, SNP_GUEST_ERR(SNP_GUEST_VMM_ERR_BUSY, fw_err));
> +		ret = 1;
> +		goto out_unlock;
> +	}
>  	if (ret && !fw_err)
>  		goto out_unlock;
>

On Thu, Jun 5, 2025 at 12:15 PM Tom Lendacky <thomas.lendacky@amd.com> wrote:
>
> On 6/5/25 10:02, Dionna Glaze wrote:
> > The ccp driver can be overloaded even with guest request rate limits.
> > The return value of -EBUSY means that there is no firmware error to
> > report back to user space, so the guest VM would see this as
> > exitinfo2 = 0. The false success can trick the guest to update its
> > message sequence number when it shouldn't have.
>
> -EBUSY from the CCP driver is an error, not a throttling condition. Either

Ah, okay thanks Tom. I'll drop it for v6. I'll see how the first patch
is received before cutting a new email.

-- 
-Dionna Glaze, PhD, CISSP, CCSP (she/her)