[PATCH v4 13/17] x86/VPMU: Add privileged PMU mode

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

Add support for privileged PMU mode which allows privileged domain (dom0)
profile both itself (and the hypervisor) and the guests. While this mode is on
profiling in guests is disabled.

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
---
 xen/arch/x86/hvm/vpmu.c     | 88 ++++++++++++++++++++++++++++++++-------------
 xen/arch/x86/traps.c        |  6 +++-
 xen/include/public/xenpmu.h |  3 ++
 3 files changed, 72 insertions(+), 25 deletions(-)

diff --git a/xen/arch/x86/hvm/vpmu.c b/xen/arch/x86/hvm/vpmu.c
index aead6af..214300d 100644
--- a/xen/arch/x86/hvm/vpmu.c
+++ b/xen/arch/x86/hvm/vpmu.c
@@ -87,6 +87,9 @@ int vpmu_do_wrmsr(unsigned int msr, uint64_t msr_content)
 {
     struct vpmu_struct *vpmu = vcpu_vpmu(current);
 
+    if ( (vpmu_mode & XENPMU_MODE_PRIV) && !is_control_domain(current->domain) )
+        return 0;
+
     if ( vpmu->arch_vpmu_ops && vpmu->arch_vpmu_ops->do_wrmsr )
     {
         int ret = vpmu->arch_vpmu_ops->do_wrmsr(msr, msr_content);
@@ -112,6 +115,9 @@ int vpmu_do_rdmsr(unsigned int msr, uint64_t *msr_content)
 {
     struct vpmu_struct *vpmu = vcpu_vpmu(current);
 
+    if ( (vpmu_mode & XENPMU_MODE_PRIV) && !is_control_domain(current->domain) )
+        return 0;
+
     if ( vpmu->arch_vpmu_ops && vpmu->arch_vpmu_ops->do_rdmsr )
     {
         int ret = vpmu->arch_vpmu_ops->do_rdmsr(msr, msr_content);
@@ -134,14 +140,18 @@ int vpmu_do_interrupt(struct cpu_user_regs *regs)
     struct vpmu_struct *vpmu;
 
     /* dom0 will handle this interrupt */
-    if ( v->domain->domain_id >= DOMID_FIRST_RESERVED )
+    if ( (vpmu_mode & XENPMU_MODE_PRIV) ||
+         (v->domain->domain_id >= DOMID_FIRST_RESERVED) )
         v = dom0->vcpu[smp_processor_id() % dom0->max_vcpus];
 
     vpmu = vcpu_vpmu(v);
-    if ( !is_hvm_domain(v->domain) )
+    if ( !vpmu_is_set(vpmu, VPMU_CONTEXT_ALLOCATED) )
+        return 0;
+
+    if ( !is_hvm_domain(v->domain) || (vpmu_mode & XENPMU_MODE_PRIV) )
     {
         /* PV guest or dom0 is doing system profiling */
-        const struct cpu_user_regs *gregs;
+        struct cpu_user_regs *gregs;
         int err;
 
         if ( v->arch.vpmu.xenpmu_data->pmu_flags & PMU_CACHED )
@@ -152,33 +162,62 @@ int vpmu_do_interrupt(struct cpu_user_regs *regs)
         err = vpmu->arch_vpmu_ops->arch_vpmu_save(v);
         vpmu_reset(vpmu, VPMU_CONTEXT_SAVE | VPMU_CONTEXT_LOADED);
 
-        /* Store appropriate registers in xenpmu_data */
-        if ( is_pv_32bit_domain(current->domain) )
+        if ( !is_hvm_domain(current->domain) )
         {
-            /*
-             * 32-bit dom0 cannot process Xen's addresses (which are 64 bit)
-             * and therefore we treat it the same way as a non-priviledged
-             * PV 32-bit domain.
-             */
-            struct compat_cpu_user_regs *cmp;
-
-            gregs = guest_cpu_user_regs();
-
-            cmp = (struct compat_cpu_user_regs *)
-                    &v->arch.vpmu.xenpmu_data->pmu.r.regs;
-            XLAT_cpu_user_regs(cmp, gregs);
+            uint16_t cs = (current->arch.flags & TF_kernel_mode) ? 0 : 0x3;
+
+            /* Store appropriate registers in xenpmu_data */
+            if ( is_pv_32bit_domain(current->domain) )
+            {
+                gregs = guest_cpu_user_regs();
+
+                if ( (vpmu_mode & XENPMU_MODE_PRIV) &&
+                     !is_pv_32bit_domain(v->domain) )
+                    memcpy(&v->arch.vpmu.xenpmu_data->pmu.r.regs,
+                           gregs, sizeof(struct cpu_user_regs));
+                else 
+                {
+                    /*
+                     * 32-bit dom0 cannot process Xen's addresses (which are
+                     * 64 bit) and therefore we treat it the same way as a
+                     * non-priviledged PV 32-bit domain.
+                     */
+
+                    struct compat_cpu_user_regs *cmp;
+
+                    cmp = (struct compat_cpu_user_regs *)
+                        &v->arch.vpmu.xenpmu_data->pmu.r.regs;
+                    XLAT_cpu_user_regs(cmp, gregs);
+                }
+            }
+            else if ( !is_control_domain(current->domain) &&
+                      !is_idle_vcpu(current) )
+            {
+                /* PV guest */
+                gregs = guest_cpu_user_regs();
+                memcpy(&v->arch.vpmu.xenpmu_data->pmu.r.regs,
+                       gregs, sizeof(struct cpu_user_regs));
+            }
+            else
+                memcpy(&v->arch.vpmu.xenpmu_data->pmu.r.regs,
+                       regs, sizeof(struct cpu_user_regs));
+
+            gregs = &v->arch.vpmu.xenpmu_data->pmu.r.regs;
+            gregs->cs = cs;
         }
-        else if ( !is_control_domain(current->domain) &&
-                 !is_idle_vcpu(current) )
+        else
         {
-            /* PV guest */
+            /* HVM guest */
+            struct segment_register cs;
+
             gregs = guest_cpu_user_regs();
             memcpy(&v->arch.vpmu.xenpmu_data->pmu.r.regs,
                    gregs, sizeof(struct cpu_user_regs));
+
+            hvm_get_segment_register(current, x86_seg_cs, &cs);
+            gregs = &v->arch.vpmu.xenpmu_data->pmu.r.regs;
+            gregs->cs = cs.attr.fields.dpl;
         }
-        else
-            memcpy(&v->arch.vpmu.xenpmu_data->pmu.r.regs,
-                   regs, sizeof(struct cpu_user_regs));
 
         v->arch.vpmu.xenpmu_data->domain_id = current->domain->domain_id;
         v->arch.vpmu.xenpmu_data->vcpu_id = current->vcpu_id;
@@ -444,7 +483,8 @@ long do_xenpmu_op(int op, XEN_GUEST_HANDLE_PARAM(xen_pmu_params_t) arg)
             return -EFAULT;
 
         mode = (uint32_t)pmu_params.d.val & XENPMU_MODE_MASK;
-        if ( mode & ~XENPMU_MODE_ON )
+        if ( (mode & ~(XENPMU_MODE_ON | XENPMU_MODE_PRIV)) ||
+             ((mode & XENPMU_MODE_ON) && (mode & XENPMU_MODE_PRIV)) )
             return -EINVAL;
 
         vpmu_mode &= ~XENPMU_MODE_MASK;
diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
index 7ff8401..1854230 100644
--- a/xen/arch/x86/traps.c
+++ b/xen/arch/x86/traps.c
@@ -2510,7 +2510,11 @@ static int emulate_privileged_op(struct cpu_user_regs *regs)
         case MSR_CORE_PERF_FIXED_CTR_CTRL...MSR_CORE_PERF_GLOBAL_OVF_CTRL:
         case MSR_AMD_FAM15H_EVNTSEL0...MSR_AMD_FAM15H_PERFCTR5:
             if ( !vpmu_do_wrmsr(regs->ecx, msr_content) )
-                goto invalid;
+            {
+                if ( (vpmu_mode & XENPMU_MODE_PRIV) &&
+                      is_control_domain(v->domain) )
+                    goto invalid;
+            }
             break;
         default:
             if ( wrmsr_hypervisor_regs(regs->ecx, msr_content) == 1 )
diff --git a/xen/include/public/xenpmu.h b/xen/include/public/xenpmu.h
index df85209..f715f30 100644
--- a/xen/include/public/xenpmu.h
+++ b/xen/include/public/xenpmu.h
@@ -56,11 +56,14 @@ DEFINE_XEN_GUEST_HANDLE(xen_pmu_params_t);
  * - XENPMU_MODE_OFF:   No PMU virtualization
  * - XENPMU_MODE_ON:    Guests can profile themselves, dom0 profiles
  *                      itself and Xen
+ * - XENPMU_MODE_PRIV:  Only dom0 has access to VPMU and it profiles
+ *                      everyone: itself, the hypervisor and the guests.
  */
 #define XENPMU_FEATURE_SHIFT      16
 #define XENPMU_MODE_MASK          ((1U << XENPMU_FEATURE_SHIFT) - 1)
 #define XENPMU_MODE_OFF           0
 #define XENPMU_MODE_ON            (1<<0)
+#define XENPMU_MODE_PRIV          (1<<1)
 
 /*
  * PMU features:
-- 
1.8.1.4