[PATCH 3/4] nested vmx: optimize for bulk access of virtual VMCS

[Dongxiao Xu <dongxiao.xu@intel.com>]

After we use the VMREAD/VMWRITE to build up the virtual VMCS, each
access to the virtual VMCS needs two VMPTRLD and one VMCLEAR to
switch the environment, which might be an overhead to performance.
This commit tries to handle multiple virtual VMCS access together
to improve the performance.

Signed-off-by: Dongxiao Xu <dongxiao.xu@intel.com>
---
 xen/arch/x86/hvm/vmx/vvmx.c |   89 +++++++++++++++++++++++++++++++++++--------
 1 files changed, 73 insertions(+), 16 deletions(-)

diff --git a/xen/arch/x86/hvm/vmx/vvmx.c b/xen/arch/x86/hvm/vmx/vvmx.c
index 2f0076a..9aba89e 100644
--- a/xen/arch/x86/hvm/vmx/vvmx.c
+++ b/xen/arch/x86/hvm/vmx/vvmx.c
@@ -829,6 +829,34 @@ static void vvmcs_to_shadow(void *vvmcs, unsigned int field)
     __vmwrite(field, value);
 }
 
+static void vvmcs_to_shadow_bulk(void *vvmcs, int n, u16 *field)
+{
+    u64 *value = NULL;
+    int i = 0;
+
+    if ( cpu_has_vmx_vmcs_shadowing )
+    {
+        value = xzalloc_array(u64, n);
+        if ( !value )
+            goto fallback;
+
+        virtual_vmcs_enter(vvmcs);
+        for ( i = 0; i < n; i++ )
+            value[i] = __vmread(field[i]);
+        virtual_vmcs_exit(vvmcs);
+
+        for ( i = 0; i < n; i++ )
+            __vmwrite(field[i], value[i]);
+
+        xfree(value);
+        return;
+    }
+
+fallback:
+    for ( i = 0; i < n; i++ )
+        vvmcs_to_shadow(vvmcs, field[i]);
+}
+
 static void shadow_to_vvmcs(void *vvmcs, unsigned int field)
 {
     u64 value;
@@ -839,6 +867,34 @@ static void shadow_to_vvmcs(void *vvmcs, unsigned int field)
         __set_vvmcs(vvmcs, field, value);
 }
 
+static void shadow_to_vvmcs_bulk(void *vvmcs, int n, u16 *field)
+{
+    u64 *value = NULL;
+    int i = 0;
+
+    if ( cpu_has_vmx_vmcs_shadowing )
+    {
+        value = xzalloc_array(u64, n);
+        if ( !value )
+            goto fallback;
+
+        for ( i = 0; i < n; i++ )
+            value[i] = __vmread(field[i]);
+
+        virtual_vmcs_enter(vvmcs);
+        for ( i = 0; i < n; i++ )
+            __vmwrite(field[i], value[i]);
+        virtual_vmcs_exit(vvmcs);
+
+        xfree(value);
+        return;
+    }
+
+fallback:
+    for ( i = 0; i < n; i++ )
+        shadow_to_vvmcs(vvmcs, field[i]);
+}
+
 static void load_shadow_control(struct vcpu *v)
 {
     /*
@@ -862,13 +918,17 @@ static void load_shadow_guest_state(struct vcpu *v)
 {
     struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v);
     void *vvmcs = nvcpu->nv_vvmcx;
-    int i;
     u32 control;
     u64 cr_gh_mask, cr_read_shadow;
 
+    u16 vmentry_fields[] = {
+        VM_ENTRY_INTR_INFO,
+        VM_ENTRY_EXCEPTION_ERROR_CODE,
+        VM_ENTRY_INSTRUCTION_LEN,
+    };
+
     /* vvmcs.gstate to shadow vmcs.gstate */
-    for ( i = 0; i < ARRAY_SIZE(vmcs_gstate_field); i++ )
-        vvmcs_to_shadow(vvmcs, vmcs_gstate_field[i]);
+    vvmcs_to_shadow_bulk(vvmcs, ARRAY_SIZE(vmcs_gstate_field), (u16 *)vmcs_gstate_field);
 
     hvm_set_cr0(__get_vvmcs(vvmcs, GUEST_CR0));
     hvm_set_cr4(__get_vvmcs(vvmcs, GUEST_CR4));
@@ -882,9 +942,7 @@ static void load_shadow_guest_state(struct vcpu *v)
 
     hvm_funcs.set_tsc_offset(v, v->arch.hvm_vcpu.cache_tsc_offset);
 
-    vvmcs_to_shadow(vvmcs, VM_ENTRY_INTR_INFO);
-    vvmcs_to_shadow(vvmcs, VM_ENTRY_EXCEPTION_ERROR_CODE);
-    vvmcs_to_shadow(vvmcs, VM_ENTRY_INSTRUCTION_LEN);
+    vvmcs_to_shadow_bulk(vvmcs, ARRAY_SIZE(vmentry_fields), vmentry_fields);
 
     /*
      * While emulate CR0 and CR4 for nested virtualization, set the CR0/CR4
@@ -904,10 +962,13 @@ static void load_shadow_guest_state(struct vcpu *v)
     if ( nvmx_ept_enabled(v) && hvm_pae_enabled(v) &&
          (v->arch.hvm_vcpu.guest_efer & EFER_LMA) )
     {
-        vvmcs_to_shadow(vvmcs, GUEST_PDPTR0);
-        vvmcs_to_shadow(vvmcs, GUEST_PDPTR1);
-        vvmcs_to_shadow(vvmcs, GUEST_PDPTR2);
-        vvmcs_to_shadow(vvmcs, GUEST_PDPTR3);
+        u16 gpdptr_fields[] = {
+            GUEST_PDPTR0,
+            GUEST_PDPTR1,
+            GUEST_PDPTR2,
+            GUEST_PDPTR3,
+        };
+        vvmcs_to_shadow_bulk(vvmcs, ARRAY_SIZE(gpdptr_fields), gpdptr_fields);
     }
 
     /* TODO: CR3 target control */
@@ -998,13 +1059,11 @@ static void virtual_vmentry(struct cpu_user_regs *regs)
 
 static void sync_vvmcs_guest_state(struct vcpu *v, struct cpu_user_regs *regs)
 {
-    int i;
     struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v);
     void *vvmcs = nvcpu->nv_vvmcx;
 
     /* copy shadow vmcs.gstate back to vvmcs.gstate */
-    for ( i = 0; i < ARRAY_SIZE(vmcs_gstate_field); i++ )
-        shadow_to_vvmcs(vvmcs, vmcs_gstate_field[i]);
+    shadow_to_vvmcs_bulk(vvmcs, ARRAY_SIZE(vmcs_gstate_field), (u16 *)vmcs_gstate_field);
     /* RIP, RSP are in user regs */
     __set_vvmcs(vvmcs, GUEST_RIP, regs->eip);
     __set_vvmcs(vvmcs, GUEST_RSP, regs->esp);
@@ -1016,13 +1075,11 @@ static void sync_vvmcs_guest_state(struct vcpu *v, struct cpu_user_regs *regs)
 
 static void sync_vvmcs_ro(struct vcpu *v)
 {
-    int i;
     struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v);
     struct nestedvmx *nvmx = &vcpu_2_nvmx(v);
     void *vvmcs = nvcpu->nv_vvmcx;
 
-    for ( i = 0; i < ARRAY_SIZE(vmcs_ro_field); i++ )
-        shadow_to_vvmcs(nvcpu->nv_vvmcx, vmcs_ro_field[i]);
+    shadow_to_vvmcs_bulk(vvmcs, ARRAY_SIZE(vmcs_ro_field), (u16 *)vmcs_ro_field);
 
     /* Adjust exit_reason/exit_qualifciation for violation case */
     if ( __get_vvmcs(vvmcs, VM_EXIT_REASON) == EXIT_REASON_EPT_VIOLATION )
-- 
1.7.1