[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH v8 16/19] x86/VPMU: Add privileged PMU mode



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@xxxxxxxxxx>
Acked-by: Kevin Tian <kevin.tian@xxxxxxxxx>
Reviewed-by: Dietmar Hahn <dietmar.hahn@xxxxxxxxxxxxxx>
Tested-by: Dietmar Hahn <dietmar.hahn@xxxxxxxxxxxxxx>
---
 xen/arch/x86/domain.c    |  4 +-
 xen/arch/x86/hvm/vpmu.c  | 99 ++++++++++++++++++++++++++++++++++--------------
 xen/arch/x86/traps.c     | 11 ++++++
 xen/include/public/pmu.h |  7 +++-
 4 files changed, 89 insertions(+), 32 deletions(-)

diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c
index bb759dd..a810d1c 100644
--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -1482,7 +1482,7 @@ void context_switch(struct vcpu *prev, struct vcpu *next)
 
     if ( is_hvm_vcpu(prev) )
     {
-        if ( (prev != next) && (vpmu_mode & XENPMU_MODE_ON) )
+        if ( (prev != next) && (vpmu_mode & XENPMU_MODE_SELF) )
             vpmu_save(prev);
 
         if ( !list_empty(&prev->arch.hvm_vcpu.tm_list) )
@@ -1526,7 +1526,7 @@ void context_switch(struct vcpu *prev, struct vcpu *next)
                            !is_hardware_domain(next->domain));
     }
 
-    if ( is_hvm_vcpu(next) && (prev != next) && (vpmu_mode & XENPMU_MODE_ON) )
+    if ( is_hvm_vcpu(next) && (prev != next) && (vpmu_mode & XENPMU_MODE_SELF) 
)
         /* Must be done with interrupts enabled */
         vpmu_load(next);
 
diff --git a/xen/arch/x86/hvm/vpmu.c b/xen/arch/x86/hvm/vpmu.c
index dacfd81..87559fe 100644
--- a/xen/arch/x86/hvm/vpmu.c
+++ b/xen/arch/x86/hvm/vpmu.c
@@ -73,7 +73,7 @@ static void __init parse_vpmu_param(char *s)
         }
         /* fall through */
     case 1:
-        vpmu_mode = XENPMU_MODE_ON;
+        vpmu_mode = XENPMU_MODE_SELF;
         break;
     }
 }
@@ -96,7 +96,9 @@ int vpmu_do_msr(unsigned int msr, uint64_t *msr_content, 
uint8_t rw)
     struct vcpu *v = current;
     struct vpmu_struct *vpmu = vcpu_vpmu(v);
 
-    if ( !(vpmu_mode & XENPMU_MODE_ON) )
+    if ( (vpmu_mode == XENPMU_MODE_OFF) ||
+         ((vpmu_mode & XENPMU_MODE_ALL) &&
+          !is_hardware_domain(current->domain)) )
         return 0;
 
     ASSERT((rw == VPMU_MSR_READ) || (rw == VPMU_MSR_WRITE));
@@ -154,8 +156,12 @@ int vpmu_do_interrupt(struct cpu_user_regs *regs)
     struct vcpu *curr_vcpu, *v = current;
     struct vpmu_struct *vpmu;
 
-    /* dom0 will handle interrupt for special domains (e.g. idle domain) */
-    if ( v->domain->domain_id >= DOMID_FIRST_RESERVED )
+    /*
+     * dom0 will handle interrupt for special domains (e.g. idle domain) or,
+     * in XENPMU_MODE_ALL, for everyone.
+     */
+    if ( (vpmu_mode & XENPMU_MODE_ALL) ||
+         (v->domain->domain_id >= DOMID_FIRST_RESERVED) )
     {
         v = choose_hwdom_vcpu();
         if ( !v )
@@ -167,7 +173,7 @@ int vpmu_do_interrupt(struct cpu_user_regs *regs)
         curr_vcpu = v;
 
     vpmu = vcpu_vpmu(v);
-    if ( !is_hvm_domain(v->domain) )
+    if ( !is_hvm_domain(v->domain) || (vpmu_mode & XENPMU_MODE_ALL) )
     {
         /* PV(H) guest or dom0 is doing system profiling */
         const struct cpu_user_regs *gregs;
@@ -179,6 +185,7 @@ int vpmu_do_interrupt(struct cpu_user_regs *regs)
             return 1;
 
         if ( is_pvh_domain(curr_vcpu->domain) &&
+             !(vpmu_mode & XENPMU_MODE_ALL) &&
              !vpmu->arch_vpmu_ops->do_interrupt(regs) )
             return 0;
 
@@ -187,34 +194,68 @@ int vpmu_do_interrupt(struct cpu_user_regs *regs)
         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(curr_vcpu->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 = (void *)&vpmu->xenpmu_data->pmu.r.regs;
-            XLAT_cpu_user_regs(cmp, gregs);
+            /* Store appropriate registers in xenpmu_data */
+            if ( is_pv_32bit_domain(curr_vcpu->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 = (void *)&vpmu->xenpmu_data->pmu.r.regs;
+                XLAT_cpu_user_regs(cmp, gregs);
+
+                /* Adjust RPL for kernel mode */
+                if ( (cmp->cs & 3) == 1 )
+                    cmp->cs &= ~3;
+            }
+            else if ( !is_hardware_domain(curr_vcpu->domain) &&
+                      !is_idle_vcpu(curr_vcpu) )
+            {
+                /* 64-bit unprivileged PV(H) guest */
+                gregs = guest_cpu_user_regs();
+                memcpy(&vpmu->xenpmu_data->pmu.r.regs,
+                       gregs, sizeof(struct cpu_user_regs));
+            }
+            else
+                memcpy(&vpmu->xenpmu_data->pmu.r.regs,
+                       regs, sizeof(struct cpu_user_regs));
+
+            if ( !is_pvh_domain(current->domain) )
+            {
+                if ( current->arch.flags & TF_kernel_mode )
+                    v->arch.vpmu.xenpmu_data->pmu.r.regs.cs &= ~3;
+            }
+            else
+            {
+                struct segment_register seg_cs;
+
+                hvm_get_segment_register(current, x86_seg_cs, &seg_cs);
+                v->arch.vpmu.xenpmu_data->pmu.r.regs.cs = seg_cs.sel;
+            }
         }
-        else if ( !is_hardware_domain(curr_vcpu->domain) &&
-                  !is_idle_vcpu(curr_vcpu) )
+        else
         {
-            /* PV(H) guest */
+            /* HVM guest */
+            struct segment_register seg_cs;
+
             gregs = guest_cpu_user_regs();
-            memcpy(&vpmu->xenpmu_data->pmu.r.regs,
+            memcpy(&v->arch.vpmu.xenpmu_data->pmu.r.regs,
                    gregs, sizeof(struct cpu_user_regs));
+
+            hvm_get_segment_register(current, x86_seg_cs, &seg_cs);
+            v->arch.vpmu.xenpmu_data->pmu.r.regs.cs = seg_cs.sel;
         }
-        else
-            memcpy(&vpmu->xenpmu_data->pmu.r.regs,
-                   regs, sizeof(struct cpu_user_regs));
 
-        vpmu->xenpmu_data->domain_id = curr_vcpu->domain->domain_id;
+        vpmu->xenpmu_data->domain_id = (curr_vcpu == v) ?
+                                       DOMID_SELF :
+                                       curr_vcpu->domain->domain_id;
         vpmu->xenpmu_data->vcpu_id = curr_vcpu->vcpu_id;
         vpmu->xenpmu_data->pcpu_id = smp_processor_id();
 
@@ -562,7 +603,9 @@ long do_xenpmu_op(int op, 
XEN_GUEST_HANDLE_PARAM(xen_pmu_params_t) arg)
         if ( copy_from_guest(&pmu_params, arg, 1) )
             return -EFAULT;
 
-        if ( pmu_params.val & ~XENPMU_MODE_ON )
+        if ( (pmu_params.val & ~(XENPMU_MODE_SELF | XENPMU_MODE_ALL)) ||
+             ((pmu_params.val & XENPMU_MODE_SELF) &&
+              (pmu_params.val & XENPMU_MODE_ALL)) )
             return -EINVAL;
 
         if ( !spin_trylock(&xenpmu_mode_lock) )
@@ -571,7 +614,7 @@ long do_xenpmu_op(int op, 
XEN_GUEST_HANDLE_PARAM(xen_pmu_params_t) arg)
         current_mode = vpmu_mode;
         vpmu_mode = pmu_params.val;
 
-        if ( vpmu_mode == XENPMU_MODE_OFF )
+        if ( (vpmu_mode == XENPMU_MODE_OFF) || (vpmu_mode == XENPMU_MODE_ALL) )
         {
             /*
              * Make sure all (non-dom0) VCPUs have unloaded their VPMUs. This
diff --git a/xen/arch/x86/traps.c b/xen/arch/x86/traps.c
index 5cd7102..33bdcf0 100644
--- a/xen/arch/x86/traps.c
+++ b/xen/arch/x86/traps.c
@@ -2573,6 +2573,10 @@ static int emulate_privileged_op(struct cpu_user_regs 
*regs)
         case MSR_AMD_FAM15H_EVNTSEL0...MSR_AMD_FAM15H_PERFCTR5:
             if ( vpmu_msr || (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) )
             {
+                if ( (vpmu_mode & XENPMU_MODE_ALL) &&
+                     !is_hardware_domain(v->domain) )
+                    break;
+
                 if ( vpmu_do_msr(regs->ecx, &msr_content, VPMU_MSR_WRITE) )
                     goto fail;
                 break;
@@ -2693,6 +2697,13 @@ static int emulate_privileged_op(struct cpu_user_regs 
*regs)
         case MSR_AMD_FAM15H_EVNTSEL0...MSR_AMD_FAM15H_PERFCTR5:
             if ( vpmu_msr || (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) )
             {
+                if ( (vpmu_mode & XENPMU_MODE_ALL) &&
+                     !is_hardware_domain(v->domain) )
+                {
+                    /* Don't leak PMU MSRs to unprivileged domains */
+                    regs->eax = regs->edx = 0;
+                    break;
+                }
                 if ( vpmu_do_msr(regs->ecx, &msr_content, VPMU_MSR_READ) )
                     goto fail;
 
diff --git a/xen/include/public/pmu.h b/xen/include/public/pmu.h
index ba0abbd..3023e52 100644
--- a/xen/include/public/pmu.h
+++ b/xen/include/public/pmu.h
@@ -48,11 +48,14 @@ DEFINE_XEN_GUEST_HANDLE(xen_pmu_params_t);
 
 /* PMU modes:
  * - XENPMU_MODE_OFF:   No PMU virtualization
- * - XENPMU_MODE_ON:    Guests can profile themselves, dom0 profiles
+ * - XENPMU_MODE_SELF:  Guests can profile themselves, dom0 profiles
  *                      itself and Xen
+ * - XENPMU_MODE_ALL:   Only dom0 has access to VPMU and it profiles
+ *                      everyone: itself, the hypervisor and the guests.
  */
 #define XENPMU_MODE_OFF           0
-#define XENPMU_MODE_ON            (1<<0)
+#define XENPMU_MODE_SELF          (1<<0)
+#define XENPMU_MODE_ALL           (1<<1)
 
 /*
  * PMU features:
-- 
1.8.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.