i386/tdx: Fetch and validate CPUID of TD guest

Use KVM_TDX_GET_CPUID to get the CPUIDs that are managed and enfored
by TDX module for TD guest. Check QEMU's configuration against the
fetched data.

Print wanring  message when 1. a feature is not supported but requested
by QEMU or 2. QEMU doesn't want to expose a feature while it is enforced
enabled.

- If cpu->enforced_cpuid is not set, prints the warning message of both
1) and 2) and tweak QEMU's configuration.

- If cpu->enforced_cpuid is set, quit if any case of 1) or 2).

Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
Link: https://lore.kernel.org/r/20250508150002.689633-52-xiaoyao.li@intel.com
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

diff --git a/target/i386/cpu.c b/target/i386/cpu.c
index be3812973f6adfa7217f4d6fd8f070a715d1000d..34364cf96ae14a95833c4b4f21f520f28a677ffd 100644 (file)
--- a/target/i386/cpu.c
+++ b/target/i386/cpu.c
@@ -5808,8 +5808,8 @@ static bool x86_cpu_have_filtered_features(X86CPU *cpu)
     return false;
 }
 
-static void mark_unavailable_features(X86CPU *cpu, FeatureWord w, uint64_t mask,
-                                      const char *verbose_prefix)
+void mark_unavailable_features(X86CPU *cpu, FeatureWord w, uint64_t mask,
+                               const char *verbose_prefix)
 {
     CPUX86State *env = &cpu->env;
     FeatureWordInfo *f = &feature_word_info[w];
@@ -5836,6 +5836,35 @@ static void mark_unavailable_features(X86CPU *cpu, FeatureWord w, uint64_t mask,
     }
 }
 
+void mark_forced_on_features(X86CPU *cpu, FeatureWord w, uint64_t mask,
+                             const char *verbose_prefix)
+{
+    CPUX86State *env = &cpu->env;
+    FeatureWordInfo *f = &feature_word_info[w];
+    int i;
+
+    if (!cpu->force_features) {
+        env->features[w] |= mask;
+    }
+
+    cpu->forced_on_features[w] |= mask;
+
+    if (!verbose_prefix) {
+        return;
+    }
+
+    for (i = 0; i < 64; ++i) {
+        if ((1ULL << i) & mask) {
+            g_autofree char *feat_word_str = feature_word_description(f);
+            warn_report("%s: %s%s%s [bit %d]",
+                        verbose_prefix,
+                        feat_word_str,
+                        f->feat_names[i] ? "." : "",
+                        f->feat_names[i] ? f->feat_names[i] : "", i);
+        }
+    }
+}
+
 static void x86_cpuid_version_get_family(Object *obj, Visitor *v,
                                          const char *name, void *opaque,
                                          Error **errp)

diff --git a/target/i386/cpu.h b/target/i386/cpu.h
index 8a4b4217d02169e0da2df7bcd21494acb8f73851..22e82444aece2d3c7fda3b4aa7b6b2bb68ef3716 100644 (file)
--- a/target/i386/cpu.h
+++ b/target/i386/cpu.h
@@ -2215,6 +2215,9 @@ struct ArchCPU {
     /* Features that were filtered out because of missing host capabilities */
     FeatureWordArray filtered_features;
 
+    /* Features that are forced enabled by underlying hypervisor, e.g., TDX */
+    FeatureWordArray forced_on_features;
+
     /* Enable PMU CPUID bits. This can't be enabled by default yet because
      * it doesn't have ABI stability guarantees, as it passes all PMU CPUID
      * bits returned by GET_SUPPORTED_CPUID (that depend on host CPU and kernel
@@ -2526,6 +2529,10 @@ void host_cpuid(uint32_t function, uint32_t count,
                 uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx);
 bool cpu_has_x2apic_feature(CPUX86State *env);
 bool is_feature_word_cpuid(uint32_t feature, uint32_t index, int reg);
+void mark_unavailable_features(X86CPU *cpu, FeatureWord w, uint64_t mask,
+                               const char *verbose_prefix);
+void mark_forced_on_features(X86CPU *cpu, FeatureWord w, uint64_t mask,
+                             const char *verbose_prefix);
 
 static inline bool x86_has_cpuid_0x1f(X86CPU *cpu)
 {

diff --git a/target/i386/kvm/tdx.c b/target/i386/kvm/tdx.c
index e35983ad9bdeae08979ccaaf5af0745175ede649..e474abf3a60901219c9d3717079c67247608096e 100644 (file)
--- a/target/i386/kvm/tdx.c
+++ b/target/i386/kvm/tdx.c
@@ -766,6 +766,106 @@ static uint32_t tdx_adjust_cpuid_features(X86ConfidentialGuest *cg,
     return value;
 }
 
+static struct kvm_cpuid2 *tdx_fetch_cpuid(CPUState *cpu, int *ret)
+{
+    struct kvm_cpuid2 *fetch_cpuid;
+    int size = KVM_MAX_CPUID_ENTRIES;
+    Error *local_err = NULL;
+    int r;
+
+    do {
+        error_free(local_err);
+        local_err = NULL;
+
+        fetch_cpuid = g_malloc0(sizeof(*fetch_cpuid) +
+                                sizeof(struct kvm_cpuid_entry2) * size);
+        fetch_cpuid->nent = size;
+        r = tdx_vcpu_ioctl(cpu, KVM_TDX_GET_CPUID, 0, fetch_cpuid, &local_err);
+        if (r == -E2BIG) {
+            g_free(fetch_cpuid);
+            size = fetch_cpuid->nent;
+        }
+    } while (r == -E2BIG);
+
+    if (r < 0) {
+        error_report_err(local_err);
+        *ret = r;
+        return NULL;
+    }
+
+    return fetch_cpuid;
+}
+
+static int tdx_check_features(X86ConfidentialGuest *cg, CPUState *cs)
+{
+    uint64_t actual, requested, unavailable, forced_on;
+    g_autofree struct kvm_cpuid2 *fetch_cpuid;
+    const char *forced_on_prefix = NULL;
+    const char *unav_prefix = NULL;
+    struct kvm_cpuid_entry2 *entry;
+    X86CPU *cpu = X86_CPU(cs);
+    CPUX86State *env = &cpu->env;
+    FeatureWordInfo *wi;
+    FeatureWord w;
+    bool mismatch = false;
+    int r;
+
+    fetch_cpuid = tdx_fetch_cpuid(cs, &r);
+    if (!fetch_cpuid) {
+        return r;
+    }
+
+    if (cpu->check_cpuid || cpu->enforce_cpuid) {
+        unav_prefix = "TDX doesn't support requested feature";
+        forced_on_prefix = "TDX forcibly sets the feature";
+    }
+
+    for (w = 0; w < FEATURE_WORDS; w++) {
+        wi = &feature_word_info[w];
+        actual = 0;
+
+        switch (wi->type) {
+        case CPUID_FEATURE_WORD:
+            entry = cpuid_find_entry(fetch_cpuid, wi->cpuid.eax, wi->cpuid.ecx);
+            if (!entry) {
+                /*
+                 * If KVM doesn't report it means it's totally configurable
+                 * by QEMU
+                 */
+                continue;
+            }
+
+            actual = cpuid_entry_get_reg(entry, wi->cpuid.reg);
+            break;
+        case MSR_FEATURE_WORD:
+            /*
+             * TODO:
+             * validate MSR features when KVM has interface report them.
+             */
+            continue;
+        }
+
+        requested = env->features[w];
+        unavailable = requested & ~actual;
+        mark_unavailable_features(cpu, w, unavailable, unav_prefix);
+        if (unavailable) {
+            mismatch = true;
+        }
+
+        forced_on = actual & ~requested;
+        mark_forced_on_features(cpu, w, forced_on, forced_on_prefix);
+        if (forced_on) {
+            mismatch = true;
+        }
+    }
+
+    if (cpu->enforce_cpuid && mismatch) {
+        return -EINVAL;
+    }
+
+    return 0;
+}
+
 static int tdx_validate_attributes(TdxGuest *tdx, Error **errp)
 {
     if ((tdx->attributes & ~tdx_caps->supported_attrs)) {
@@ -1161,4 +1261,5 @@ static void tdx_guest_class_init(ObjectClass *oc, const void *data)
     x86_klass->kvm_type = tdx_kvm_type;
     x86_klass->cpu_instance_init = tdx_cpu_instance_init;
     x86_klass->adjust_cpuid_features = tdx_adjust_cpuid_features;
+    x86_klass->check_features = tdx_check_features;
 }