[PATCH 2/3] x86: Add support for CpuidUserDis

[Alejandro Vallejo <alejandro.vallejo@xxxxxxxxx>]

Includes a refactor to move vendor-specific probes to vendor-specific
files. Furthermore, because CpuIdUserDis is reported in Cpuid itself,
the extended leaf containing that bit must be retrieved before calling
c_early_init()

Signed-off-by: Alejandro Vallejo <alejandro.vallejo@xxxxxxxxx>
---
 xen/arch/x86/cpu/amd.c         | 29 ++++++++++++++++++-
 xen/arch/x86/cpu/common.c      | 51 ++++++++++++++++++----------------
 xen/arch/x86/cpu/intel.c       | 11 +++++++-
 xen/arch/x86/include/asm/amd.h |  1 +
 4 files changed, 66 insertions(+), 26 deletions(-)

diff --git a/xen/arch/x86/cpu/amd.c b/xen/arch/x86/cpu/amd.c
index caafe44740..9269015edd 100644
--- a/xen/arch/x86/cpu/amd.c
+++ b/xen/arch/x86/cpu/amd.c
@@ -271,8 +271,20 @@ static void __init noinline amd_init_levelling(void)
 {
        const struct cpuidmask *m = NULL;
 
-       if (probe_cpuid_faulting())
+       /*
+        * If there's support for CpuidUserDis or CPUID faulting then
+        * we can skip levelling because CPUID accesses are trapped anyway.
+        *
+        * CPUID faulting is an Intel feature analogous to CpuidUserDis, so
+        * that can only be present when Xen is itself virtualized (because
+        * it can be emulated)
+        */
+       if ((cpu_has_hypervisor && probe_cpuid_faulting()) ||
+           boot_cpu_has(X86_FEATURE_CPUID_USER_DIS)) {
+               expected_levelling_cap |= LCAP_faulting;
+               levelling_caps |=  LCAP_faulting;
                return;
+       }
 
        probe_masking_msrs();
 
@@ -363,6 +375,21 @@ static void __init noinline amd_init_levelling(void)
                ctxt_switch_masking = amd_ctxt_switch_masking;
 }
 
+void amd_set_cpuid_user_dis(bool enable)
+{
+       const uint64_t msr_addr = MSR_K8_HWCR;
+       const uint64_t bit = K8_HWCR_CPUID_USER_DIS;
+       uint64_t val;
+
+       rdmsrl(msr_addr, val);
+
+       if (!!(val & bit) == enable)
+               return;
+
+       val ^= bit;
+       wrmsrl(msr_addr, val);
+}
+
 /*
  * Check for the presence of an AMD erratum. Arguments are defined in amd.h 
  * for each known erratum. Return 1 if erratum is found.
diff --git a/xen/arch/x86/cpu/common.c b/xen/arch/x86/cpu/common.c
index edc4db1335..9bbb385db4 100644
--- a/xen/arch/x86/cpu/common.c
+++ b/xen/arch/x86/cpu/common.c
@@ -4,6 +4,7 @@
 #include <xen/param.h>
 #include <xen/smp.h>
 
+#include <asm/amd.h>
 #include <asm/cpu-policy.h>
 #include <asm/current.h>
 #include <asm/debugreg.h>
@@ -131,17 +132,6 @@ bool __init probe_cpuid_faulting(void)
        uint64_t val;
        int rc;
 
-       /*
-        * Don't bother looking for CPUID faulting if we aren't virtualised on
-        * AMD or Hygon hardware - it won't be present.  Likewise for Fam0F
-        * Intel hardware.
-        */
-       if (((boot_cpu_data.x86_vendor & (X86_VENDOR_AMD | X86_VENDOR_HYGON)) ||
-            ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
-             boot_cpu_data.x86 == 0xf)) &&
-           !cpu_has_hypervisor)
-               return false;
-
        if ((rc = rdmsr_safe(MSR_INTEL_PLATFORM_INFO, val)) == 0)
                raw_cpu_policy.platform_info.cpuid_faulting =
                        val & MSR_PLATFORM_INFO_CPUID_FAULTING;
@@ -155,8 +145,6 @@ bool __init probe_cpuid_faulting(void)
                return false;
        }
 
-       expected_levelling_cap |= LCAP_faulting;
-       levelling_caps |=  LCAP_faulting;
        setup_force_cpu_cap(X86_FEATURE_CPUID_FAULTING);
 
        return true;
@@ -179,8 +167,10 @@ static void set_cpuid_faulting(bool enable)
 void ctxt_switch_levelling(const struct vcpu *next)
 {
        const struct domain *nextd = next ? next->domain : NULL;
+       bool enable_cpuid_faulting;
 
-       if (cpu_has_cpuid_faulting) {
+       if (cpu_has_cpuid_faulting ||
+           boot_cpu_has(X86_FEATURE_CPUID_USER_DIS)) {
                /*
                 * No need to alter the faulting setting if we are switching
                 * to idle; it won't affect any code running in idle context.
@@ -201,12 +191,18 @@ void ctxt_switch_levelling(const struct vcpu *next)
                 * an interim escape hatch in the form of
                 * `dom0=no-cpuid-faulting` to restore the older behaviour.
                 */
-               set_cpuid_faulting(nextd && (opt_dom0_cpuid_faulting ||
-                                            !is_control_domain(nextd) ||
-                                            !is_pv_domain(nextd)) &&
-                                  (is_pv_domain(nextd) ||
-                                   next->arch.msrs->
-                                   misc_features_enables.cpuid_faulting));
+               enable_cpuid_faulting = nextd && (opt_dom0_cpuid_faulting ||
+                                                 !is_control_domain(nextd) ||
+                                                 !is_pv_domain(nextd)) &&
+                                       (is_pv_domain(nextd) ||
+                                        next->arch.msrs->
+                                        misc_features_enables.cpuid_faulting);
+
+               if (cpu_has_cpuid_faulting)
+                       set_cpuid_faulting(enable_cpuid_faulting);
+               else
+                       amd_set_cpuid_user_dis(enable_cpuid_faulting);
+
                return;
        }
 
@@ -415,6 +411,17 @@ static void generic_identify(struct cpuinfo_x86 *c)
        c->apicid = phys_pkg_id((ebx >> 24) & 0xFF, 0);
        c->phys_proc_id = c->apicid;
 
+       eax = cpuid_eax(0x80000000);
+       if ((eax >> 16) == 0x8000)
+               c->extended_cpuid_level = eax;
+
+       /*
+        * These AMD-defined flags are out of place, but we need
+        * them early for the CPUID faulting probe code
+        */
+       if (c->extended_cpuid_level >= 0x80000021)
+               c->x86_capability[FEATURESET_e21a] = cpuid_eax(0x80000021);
+
        if (this_cpu->c_early_init)
                this_cpu->c_early_init(c);
 
@@ -431,10 +438,6 @@ static void generic_identify(struct cpuinfo_x86 *c)
             (cpuid_ecx(CPUID_PM_LEAF) & CPUID6_ECX_APERFMPERF_CAPABILITY) )
                __set_bit(X86_FEATURE_APERFMPERF, c->x86_capability);
 
-       eax = cpuid_eax(0x80000000);
-       if ((eax >> 16) == 0x8000)
-               c->extended_cpuid_level = eax;
-
        /* AMD-defined flags: level 0x80000001 */
        if (c->extended_cpuid_level >= 0x80000001)
                cpuid(0x80000001, &tmp, &tmp,
diff --git a/xen/arch/x86/cpu/intel.c b/xen/arch/x86/cpu/intel.c
index 71fc1a1e18..7e5c657758 100644
--- a/xen/arch/x86/cpu/intel.c
+++ b/xen/arch/x86/cpu/intel.c
@@ -226,8 +226,17 @@ static void cf_check intel_ctxt_switch_masking(const 
struct vcpu *next)
  */
 static void __init noinline intel_init_levelling(void)
 {
-       if (probe_cpuid_faulting())
+       /* Intel Fam0f is old enough that probing for CPUID faulting support
+        * introduces spurious #GP(0) when the appropriate MSRs are read,
+        * so skip it altogether. In the case where Xen is virtualized these
+        * MSRs may be emulated though, so we allow it in that case.
+        */
+       if ((cpu_has_hypervisor || boot_cpu_data.x86 !=0xf) &&
+           probe_cpuid_faulting()) {
+               expected_levelling_cap |= LCAP_faulting;
+               levelling_caps |=  LCAP_faulting;
                return;
+       }
 
        probe_masking_msrs();
 
diff --git a/xen/arch/x86/include/asm/amd.h b/xen/arch/x86/include/asm/amd.h
index a975d3de26..09ee52dc1c 100644
--- a/xen/arch/x86/include/asm/amd.h
+++ b/xen/arch/x86/include/asm/amd.h
@@ -155,5 +155,6 @@ extern bool amd_legacy_ssbd;
 extern bool amd_virt_spec_ctrl;
 bool amd_setup_legacy_ssbd(void);
 void amd_set_legacy_ssbd(bool enable);
+void amd_set_cpuid_user_dis(bool enable);
 
 #endif /* __AMD_H__ */
-- 
2.34.1