[PATCH v6 10/19] x86/VPMU: Interface for setting PMU mode and flags

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

Add runtime interface for setting PMU mode and flags. Three main modes are
provided:
* PMU off
* PMU on: Guests can access PMU MSRs and receive PMU interrupts. dom0
  profiles itself and the hypervisor.
* dom0-only PMU: dom0 collects samples for both itself and guests.

For feature flags only Intel's BTS is currently supported.

Mode and flags are set via HYPERVISOR_xenpmu_op hypercall.

Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Reviewed-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
---
 xen/arch/x86/domain.c              |   4 +-
 xen/arch/x86/hvm/svm/vpmu.c        |   4 +-
 xen/arch/x86/hvm/vmx/vpmu_core2.c  |  10 +--
 xen/arch/x86/hvm/vpmu.c            | 121 ++++++++++++++++++++++++++++++++++---
 xen/arch/x86/x86_64/compat/entry.S |   4 ++
 xen/arch/x86/x86_64/entry.S        |   4 ++
 xen/include/asm-x86/hvm/vpmu.h     |  14 ++---
 xen/include/public/pmu.h           |  48 +++++++++++++++
 xen/include/public/xen.h           |   1 +
 xen/include/xen/hypercall.h        |   4 ++
 10 files changed, 188 insertions(+), 26 deletions(-)

diff --git a/xen/arch/x86/domain.c b/xen/arch/x86/domain.c
index 2a9c6fc..483e85b 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)
+        if ( (prev != next) && (vpmu_mode & XENPMU_MODE_ON) )
             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) )
+    if ( is_hvm_vcpu(next) && (prev != next) && (vpmu_mode & XENPMU_MODE_ON) )
         /* Must be done with interrupts enabled */
         vpmu_load(next);
 
diff --git a/xen/arch/x86/hvm/svm/vpmu.c b/xen/arch/x86/hvm/svm/vpmu.c
index ebdba8e..3828c4b 100644
--- a/xen/arch/x86/hvm/svm/vpmu.c
+++ b/xen/arch/x86/hvm/svm/vpmu.c
@@ -472,14 +472,14 @@ struct arch_vpmu_ops amd_vpmu_ops = {
     .arch_vpmu_dump = amd_vpmu_dump
 };
 
-int svm_vpmu_initialise(struct vcpu *v, unsigned int vpmu_flags)
+int svm_vpmu_initialise(struct vcpu *v)
 {
     struct vpmu_struct *vpmu = vcpu_vpmu(v);
     uint8_t family = current_cpu_data.x86;
     int ret = 0;
 
     /* vpmu enabled? */
-    if ( !vpmu_flags )
+    if ( vpmu_mode == XENPMU_MODE_OFF )
         return 0;
 
     switch ( family )
diff --git a/xen/arch/x86/hvm/vmx/vpmu_core2.c b/xen/arch/x86/hvm/vmx/vpmu_core2.c
index 1fe583f..8d5b26e 100644
--- a/xen/arch/x86/hvm/vmx/vpmu_core2.c
+++ b/xen/arch/x86/hvm/vmx/vpmu_core2.c
@@ -703,13 +703,13 @@ static int core2_vpmu_do_interrupt(struct cpu_user_regs *regs)
     return 1;
 }
 
-static int core2_vpmu_initialise(struct vcpu *v, unsigned int vpmu_flags)
+static int core2_vpmu_initialise(struct vcpu *v)
 {
     struct vpmu_struct *vpmu = vcpu_vpmu(v);
     u64 msr_content;
     struct cpuinfo_x86 *c = &current_cpu_data;
 
-    if ( !(vpmu_flags & VPMU_BOOT_BTS) )
+    if ( !(vpmu_features & XENPMU_FEATURE_INTEL_BTS) )
         goto func_out;
     /* Check the 'Debug Store' feature in the CPUID.EAX[1]:EDX[21] */
     if ( cpu_has(c, X86_FEATURE_DS) )
@@ -821,7 +821,7 @@ struct arch_vpmu_ops core2_no_vpmu_ops = {
     .do_cpuid = core2_no_vpmu_do_cpuid,
 };
 
-int vmx_vpmu_initialise(struct vcpu *v, unsigned int vpmu_flags)
+int vmx_vpmu_initialise(struct vcpu *v)
 {
     struct vpmu_struct *vpmu = vcpu_vpmu(v);
     uint8_t family = current_cpu_data.x86;
@@ -829,7 +829,7 @@ int vmx_vpmu_initialise(struct vcpu *v, unsigned int vpmu_flags)
     int ret = 0;
 
     vpmu->arch_vpmu_ops = &core2_no_vpmu_ops;
-    if ( !vpmu_flags )
+    if ( vpmu_mode == XENPMU_MODE_OFF )
         return 0;
 
     if ( family == 6 )
@@ -872,7 +872,7 @@ int vmx_vpmu_initialise(struct vcpu *v, unsigned int vpmu_flags)
         /* future: */
         case 0x3d:
         case 0x4e:
-            ret = core2_vpmu_initialise(v, vpmu_flags);
+            ret = core2_vpmu_initialise(v);
             if ( !ret )
                 vpmu->arch_vpmu_ops = &core2_vpmu_ops;
             return ret;
diff --git a/xen/arch/x86/hvm/vpmu.c b/xen/arch/x86/hvm/vpmu.c
index 0acc486..d3de8a6 100644
--- a/xen/arch/x86/hvm/vpmu.c
+++ b/xen/arch/x86/hvm/vpmu.c
@@ -21,6 +21,7 @@
 #include <xen/config.h>
 #include <xen/sched.h>
 #include <xen/xenoprof.h>
+#include <xen/guest_access.h>
 #include <asm/regs.h>
 #include <asm/types.h>
 #include <asm/msr.h>
@@ -38,7 +39,8 @@
  * "vpmu=off" : vpmu generally disabled
  * "vpmu=bts" : vpmu enabled and Intel BTS feature switched on.
  */
-static unsigned int __read_mostly opt_vpmu_enabled;
+uint64_t __read_mostly vpmu_mode = XENPMU_MODE_OFF;
+uint64_t __read_mostly vpmu_features = 0;
 static void parse_vpmu_param(char *s);
 custom_param("vpmu", parse_vpmu_param);
 
@@ -52,7 +54,7 @@ static void __init parse_vpmu_param(char *s)
         break;
     default:
         if ( !strcmp(s, "bts") )
-            opt_vpmu_enabled |= VPMU_BOOT_BTS;
+            vpmu_features |= XENPMU_FEATURE_INTEL_BTS;
         else if ( *s )
         {
             printk("VPMU: unknown flag: %s - vpmu disabled!\n", s);
@@ -60,7 +62,7 @@ static void __init parse_vpmu_param(char *s)
         }
         /* fall through */
     case 1:
-        opt_vpmu_enabled |= VPMU_BOOT_ENABLED;
+        vpmu_mode = XENPMU_MODE_ON;
         break;
     }
 }
@@ -77,6 +79,9 @@ int vpmu_do_wrmsr(unsigned int msr, uint64_t msr_content)
 {
     struct vpmu_struct *vpmu = vcpu_vpmu(current);
 
+    if ( !(vpmu_mode & XENPMU_MODE_ON) )
+        return 0;
+
     if ( vpmu->arch_vpmu_ops && vpmu->arch_vpmu_ops->do_wrmsr )
         return vpmu->arch_vpmu_ops->do_wrmsr(msr, msr_content);
     return 0;
@@ -86,6 +91,9 @@ int vpmu_do_rdmsr(unsigned int msr, uint64_t *msr_content)
 {
     struct vpmu_struct *vpmu = vcpu_vpmu(current);
 
+    if ( !(vpmu_mode & XENPMU_MODE_ON) )
+        return 0;
+
     if ( vpmu->arch_vpmu_ops && vpmu->arch_vpmu_ops->do_rdmsr )
         return vpmu->arch_vpmu_ops->do_rdmsr(msr, msr_content);
     return 0;
@@ -237,19 +245,19 @@ void vpmu_initialise(struct vcpu *v)
     switch ( vendor )
     {
     case X86_VENDOR_AMD:
-        if ( svm_vpmu_initialise(v, opt_vpmu_enabled) != 0 )
-            opt_vpmu_enabled = 0;
+        if ( svm_vpmu_initialise(v) != 0 )
+            vpmu_mode = XENPMU_MODE_OFF;
         break;
 
     case X86_VENDOR_INTEL:
-        if ( vmx_vpmu_initialise(v, opt_vpmu_enabled) != 0 )
-            opt_vpmu_enabled = 0;
+        if ( vmx_vpmu_initialise(v) != 0 )
+            vpmu_mode = XENPMU_MODE_OFF;
         break;
 
     default:
         printk("VPMU: Initialization failed. "
                "Unknown CPU vendor %d\n", vendor);
-        opt_vpmu_enabled = 0;
+        vpmu_mode = XENPMU_MODE_OFF;
         break;
     }
 }
@@ -271,3 +279,100 @@ void vpmu_dump(struct vcpu *v)
         vpmu->arch_vpmu_ops->arch_vpmu_dump(v);
 }
 
+/* Unload VPMU contexts */
+static void vpmu_unload_all(void)
+{
+    struct domain *d;
+    struct vcpu *v;
+    struct vpmu_struct *vpmu;
+
+    for_each_domain(d)
+    {
+        for_each_vcpu ( d, v )
+        {
+            if ( v != current )
+                vcpu_pause(v);
+            vpmu = vcpu_vpmu(v);
+
+            if ( !vpmu_is_set(vpmu, VPMU_CONTEXT_LOADED) )
+            {
+                if ( v != current )
+                    vcpu_unpause(v);
+                continue;
+            }
+
+            vpmu_set(vpmu, VPMU_CONTEXT_SAVE);
+            on_selected_cpus(cpumask_of(vpmu->last_pcpu),
+                             vpmu_save_force, (void *)v, 1);
+            vpmu_reset(vpmu, VPMU_CONTEXT_LOADED);
+
+            if ( v != current )
+                vcpu_unpause(v);
+        }
+    }
+}
+
+
+long do_xenpmu_op(int op, XEN_GUEST_HANDLE_PARAM(xen_pmu_params_t) arg)
+{
+    int ret = -EINVAL;
+    xen_pmu_params_t pmu_params;
+
+    switch ( op )
+    {
+    case XENPMU_mode_set:
+        if ( !is_control_domain(current->domain) )
+            return -EPERM;
+
+        if ( copy_from_guest(&pmu_params, arg, 1) )
+            return -EFAULT;
+
+        if ( pmu_params.d.val & ~XENPMU_MODE_ON )
+            return -EINVAL;
+
+        vpmu_mode = pmu_params.d.val;
+        if ( vpmu_mode == XENPMU_MODE_OFF )
+            /*
+             * After this VPMU context will never be loaded during context
+             * switch. We also prevent PMU MSR accesses (which can load
+             * context) when VPMU is disabled.
+             */
+            vpmu_unload_all();
+
+        ret = 0;
+        break;
+
+    case XENPMU_mode_get:
+        pmu_params.d.val = vpmu_mode;
+        pmu_params.v.version.maj = XENPMU_VER_MAJ;
+        pmu_params.v.version.min = XENPMU_VER_MIN;
+        if ( copy_to_guest(arg, &pmu_params, 1) )
+            return -EFAULT;
+        ret = 0;
+        break;
+
+    case XENPMU_feature_set:
+        if ( !is_control_domain(current->domain) )
+            return -EPERM;
+
+        if ( copy_from_guest(&pmu_params, arg, 1) )
+            return -EFAULT;
+
+        if ( pmu_params.d.val & ~XENPMU_FEATURE_INTEL_BTS )
+            return -EINVAL;
+
+        vpmu_features = pmu_params.d.val;
+
+        ret = 0;
+        break;
+
+    case XENPMU_feature_get:
+        pmu_params.d.val = vpmu_mode;
+        if ( copy_to_guest(arg, &pmu_params, 1) )
+            return -EFAULT;
+        ret = 0;
+        break;
+     }
+
+    return ret;
+}
diff --git a/xen/arch/x86/x86_64/compat/entry.S b/xen/arch/x86/x86_64/compat/entry.S
index 32b3bcc..e1ccf1c 100644
--- a/xen/arch/x86/x86_64/compat/entry.S
+++ b/xen/arch/x86/x86_64/compat/entry.S
@@ -416,6 +416,8 @@ ENTRY(compat_hypercall_table)
         .quad do_domctl
         .quad compat_kexec_op
         .quad do_tmem_op
+        .quad do_ni_hypercall           /* reserved for XenClient */
+        .quad do_xenpmu_op              /* 40 */
         .rept __HYPERVISOR_arch_0-((.-compat_hypercall_table)/8)
         .quad compat_ni_hypercall
         .endr
@@ -464,6 +466,8 @@ ENTRY(compat_hypercall_args_table)
         .byte 1 /* do_domctl                */
         .byte 2 /* compat_kexec_op          */
         .byte 1 /* do_tmem_op               */
+        .byte 0 /* reserved for XenClient   */
+        .byte 2 /* do_xenpmu_op             */  /* 40 */
         .rept __HYPERVISOR_arch_0-(.-compat_hypercall_args_table)
         .byte 0 /* compat_ni_hypercall      */
         .endr
diff --git a/xen/arch/x86/x86_64/entry.S b/xen/arch/x86/x86_64/entry.S
index 3ea4683..c36ffce 100644
--- a/xen/arch/x86/x86_64/entry.S
+++ b/xen/arch/x86/x86_64/entry.S
@@ -757,6 +757,8 @@ ENTRY(hypercall_table)
         .quad do_domctl
         .quad do_kexec_op
         .quad do_tmem_op
+        .quad do_ni_hypercall       /* reserved for XenClient */
+        .quad do_xenpmu_op          /* 40 */
         .rept __HYPERVISOR_arch_0-((.-hypercall_table)/8)
         .quad do_ni_hypercall
         .endr
@@ -805,6 +807,8 @@ ENTRY(hypercall_args_table)
         .byte 1 /* do_domctl            */
         .byte 2 /* do_kexec             */
         .byte 1 /* do_tmem_op           */
+        .byte 0 /* reserved for XenClient */
+        .byte 2 /* do_xenpmu_op         */  /* 40 */
         .rept __HYPERVISOR_arch_0-(.-hypercall_args_table)
         .byte 0 /* do_ni_hypercall      */
         .endr
diff --git a/xen/include/asm-x86/hvm/vpmu.h b/xen/include/asm-x86/hvm/vpmu.h
index 9e49a78..3dfdab7 100644
--- a/xen/include/asm-x86/hvm/vpmu.h
+++ b/xen/include/asm-x86/hvm/vpmu.h
@@ -24,13 +24,6 @@
 
 #include <public/pmu.h>
 
-/*
- * Flag bits given as a string on the hypervisor boot parameter 'vpmu'.
- * See arch/x86/hvm/vpmu.c.
- */
-#define VPMU_BOOT_ENABLED 0x1    /* vpmu generally enabled. */
-#define VPMU_BOOT_BTS     0x2    /* Intel BTS feature wanted. */
-
 #define vcpu_vpmu(vcpu)   (&(vcpu)->arch.vpmu)
 #define vpmu_vcpu(vpmu)   container_of(vpmu, struct vcpu, arch.vpmu)
 
@@ -58,8 +51,8 @@ struct arch_vpmu_ops {
     void (*arch_vpmu_dump)(const struct vcpu *);
 };
 
-int vmx_vpmu_initialise(struct vcpu *, unsigned int flags);
-int svm_vpmu_initialise(struct vcpu *, unsigned int flags);
+int vmx_vpmu_initialise(struct vcpu *);
+int svm_vpmu_initialise(struct vcpu *);
 
 struct vpmu_struct {
     u32 flags;
@@ -99,5 +92,8 @@ void vpmu_dump(struct vcpu *v);
 extern int acquire_pmu_ownership(int pmu_ownership);
 extern void release_pmu_ownership(int pmu_ownership);
 
+extern uint64_t vpmu_mode;
+extern uint64_t vpmu_features;
+
 #endif /* __ASM_X86_HVM_VPMU_H_*/
 
diff --git a/xen/include/public/pmu.h b/xen/include/public/pmu.h
index 3ffd2cf..f91d935 100644
--- a/xen/include/public/pmu.h
+++ b/xen/include/public/pmu.h
@@ -13,6 +13,54 @@
 #define XENPMU_VER_MAJ    0
 #define XENPMU_VER_MIN    0
 
+/*
+ * ` enum neg_errnoval
+ * ` HYPERVISOR_xenpmu_op(enum xenpmu_op cmd, struct xenpmu_params *args);
+ *
+ * @cmd  == XENPMU_* (PMU operation)
+ * @args == struct xenpmu_params
+ */
+/* ` enum xenpmu_op { */
+#define XENPMU_mode_get        0 /* Also used for getting PMU version */
+#define XENPMU_mode_set        1
+#define XENPMU_feature_get     2
+#define XENPMU_feature_set     3
+/* ` } */
+
+/* Parameters structure for HYPERVISOR_xenpmu_op call */
+struct xen_pmu_params {
+    /* IN/OUT parameters */
+    union {
+        struct version {
+            uint8_t maj;
+            uint8_t min;
+        } version;
+        uint64_t pad;
+    } v;
+    union {
+        uint64_t val;
+        XEN_GUEST_HANDLE(void) valp;
+    } d;
+
+    /* IN parameters */
+    uint64_t vcpu;
+};
+typedef struct xen_pmu_params xen_pmu_params_t;
+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
+ *                      itself and Xen
+ */
+#define XENPMU_MODE_OFF           0
+#define XENPMU_MODE_ON            (1<<0)
+
+/*
+ * PMU features:
+ * - XENPMU_FEATURE_INTEL_BTS: Intel BTS support (ignored on AMD)
+ */
+#define XENPMU_FEATURE_INTEL_BTS  1
 
 /* Shared between hypervisor and PV domain */
 struct xen_pmu_data {
diff --git a/xen/include/public/xen.h b/xen/include/public/xen.h
index a6a2092..0766790 100644
--- a/xen/include/public/xen.h
+++ b/xen/include/public/xen.h
@@ -101,6 +101,7 @@ DEFINE_XEN_GUEST_HANDLE(xen_ulong_t);
 #define __HYPERVISOR_kexec_op             37
 #define __HYPERVISOR_tmem_op              38
 #define __HYPERVISOR_xc_reserved_op       39 /* reserved for XenClient */
+#define __HYPERVISOR_xenpmu_op            40
 
 /* Architecture-specific hypercall definitions. */
 #define __HYPERVISOR_arch_0               48
diff --git a/xen/include/xen/hypercall.h b/xen/include/xen/hypercall.h
index a9e5229..cf34547 100644
--- a/xen/include/xen/hypercall.h
+++ b/xen/include/xen/hypercall.h
@@ -14,6 +14,7 @@
 #include <public/event_channel.h>
 #include <public/tmem.h>
 #include <public/version.h>
+#include <public/pmu.h>
 #include <asm/hypercall.h>
 #include <xsm/xsm.h>
 
@@ -139,6 +140,9 @@ do_tmem_op(
 extern long
 do_xenoprof_op(int op, XEN_GUEST_HANDLE_PARAM(void) arg);
 
+extern long
+do_xenpmu_op(int op, XEN_GUEST_HANDLE_PARAM(xen_pmu_params_t) arg);
+
 #ifdef CONFIG_COMPAT
 
 extern int
-- 
1.8.1.4