[Xen-devel] [PATCH 2/3] VMX: allocate VMCS pages from domain heap

["Jan Beulich" <JBeulich@xxxxxxxx>]

There being only very few uses of the virtual address of a VMCS,
convert these cases to establish a mapping and lift the Xen heap
restriction from the VMCS allocation.

Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>

--- a/xen/arch/x86/hvm/vmx/vmcs.c
+++ b/xen/arch/x86/hvm/vmx/vmcs.c
@@ -119,8 +119,8 @@ const u32 vmx_introspection_force_enable
 const unsigned int vmx_introspection_force_enabled_msrs_size =
     ARRAY_SIZE(vmx_introspection_force_enabled_msrs);
 
-static DEFINE_PER_CPU_READ_MOSTLY(struct vmcs_struct *, vmxon_region);
-static DEFINE_PER_CPU(struct vmcs_struct *, current_vmcs);
+static DEFINE_PER_CPU_READ_MOSTLY(paddr_t, vmxon_region);
+static DEFINE_PER_CPU(paddr_t, current_vmcs);
 static DEFINE_PER_CPU(struct list_head, active_vmcs_list);
 DEFINE_PER_CPU(bool_t, vmxon);
 
@@ -483,25 +483,28 @@ static int vmx_init_vmcs_config(void)
     return 0;
 }
 
-static struct vmcs_struct *vmx_alloc_vmcs(void)
+static paddr_t vmx_alloc_vmcs(void)
 {
+    struct page_info *pg;
     struct vmcs_struct *vmcs;
 
-    if ( (vmcs = alloc_xenheap_page()) == NULL )
+    if ( (pg = alloc_domheap_page(NULL, 0)) == NULL )
     {
         gdprintk(XENLOG_WARNING, "Failed to allocate VMCS.\n");
-        return NULL;
+        return 0;
     }
 
+    vmcs = __map_domain_page(pg);
     clear_page(vmcs);
     vmcs->vmcs_revision_id = vmcs_revision_id;
+    unmap_domain_page(vmcs);
 
-    return vmcs;
+    return page_to_maddr(pg);
 }
 
-static void vmx_free_vmcs(struct vmcs_struct *vmcs)
+static void vmx_free_vmcs(paddr_t pa)
 {
-    free_xenheap_page(vmcs);
+    free_domheap_page(maddr_to_page(pa));
 }
 
 static void __vmx_clear_vmcs(void *info)
@@ -514,7 +517,7 @@ static void __vmx_clear_vmcs(void *info)
 
     if ( arch_vmx->active_cpu == smp_processor_id() )
     {
-        __vmpclear(virt_to_maddr(arch_vmx->vmcs));
+        __vmpclear(arch_vmx->vmcs_pa);
         if ( arch_vmx->vmcs_shadow_maddr )
             __vmpclear(arch_vmx->vmcs_shadow_maddr);
 
@@ -523,8 +526,8 @@ static void __vmx_clear_vmcs(void *info)
 
         list_del(&arch_vmx->active_list);
 
-        if ( arch_vmx->vmcs == this_cpu(current_vmcs) )
-            this_cpu(current_vmcs) = NULL;
+        if ( arch_vmx->vmcs_pa == this_cpu(current_vmcs) )
+            this_cpu(current_vmcs) = 0;
     }
 }
 
@@ -550,8 +553,8 @@ static void vmx_load_vmcs(struct vcpu *v
 
     ASSERT(v->arch.hvm_vmx.active_cpu == smp_processor_id());
 
-    __vmptrld(virt_to_maddr(v->arch.hvm_vmx.vmcs));
-    this_cpu(current_vmcs) = v->arch.hvm_vmx.vmcs;
+    __vmptrld(v->arch.hvm_vmx.vmcs_pa);
+    this_cpu(current_vmcs) = v->arch.hvm_vmx.vmcs_pa;
 
     local_irq_restore(flags);
 }
@@ -565,11 +568,11 @@ int vmx_cpu_up_prepare(unsigned int cpu)
     if ( nvmx_cpu_up_prepare(cpu) != 0 )
         printk("CPU%d: Could not allocate virtual VMCS buffer.\n", cpu);
 
-    if ( per_cpu(vmxon_region, cpu) != NULL )
+    if ( per_cpu(vmxon_region, cpu) )
         return 0;
 
     per_cpu(vmxon_region, cpu) = vmx_alloc_vmcs();
-    if ( per_cpu(vmxon_region, cpu) != NULL )
+    if ( per_cpu(vmxon_region, cpu) )
         return 0;
 
     printk("CPU%d: Could not allocate host VMCS\n", cpu);
@@ -580,7 +583,7 @@ int vmx_cpu_up_prepare(unsigned int cpu)
 void vmx_cpu_dead(unsigned int cpu)
 {
     vmx_free_vmcs(per_cpu(vmxon_region, cpu));
-    per_cpu(vmxon_region, cpu) = NULL;
+    per_cpu(vmxon_region, cpu) = 0;
     nvmx_cpu_dead(cpu);
 }
 
@@ -638,7 +641,7 @@ int vmx_cpu_up(void)
     if ( (rc = vmx_cpu_up_prepare(cpu)) != 0 )
         return rc;
 
-    switch ( __vmxon(virt_to_maddr(this_cpu(vmxon_region))) )
+    switch ( __vmxon(this_cpu(vmxon_region)) )
     {
     case -2: /* #UD or #GP */
         if ( bios_locked &&
@@ -710,7 +713,7 @@ bool_t vmx_vmcs_try_enter(struct vcpu *v
      * vmx_vmcs_enter/exit and scheduling tail critical regions.
      */
     if ( likely(v == current) )
-        return v->arch.hvm_vmx.vmcs == this_cpu(current_vmcs);
+        return v->arch.hvm_vmx.vmcs_pa == this_cpu(current_vmcs);
 
     fv = &this_cpu(foreign_vmcs);
 
@@ -906,17 +909,17 @@ int vmx_check_msr_bitmap(unsigned long *
 /*
  * Switch VMCS between layer 1 & 2 guest
  */
-void vmx_vmcs_switch(struct vmcs_struct *from, struct vmcs_struct *to)
+void vmx_vmcs_switch(paddr_t from, paddr_t to)
 {
     struct arch_vmx_struct *vmx = &current->arch.hvm_vmx;
     spin_lock(&vmx->vmcs_lock);
 
-    __vmpclear(virt_to_maddr(from));
+    __vmpclear(from);
     if ( vmx->vmcs_shadow_maddr )
         __vmpclear(vmx->vmcs_shadow_maddr);
-    __vmptrld(virt_to_maddr(to));
+    __vmptrld(to);
 
-    vmx->vmcs = to;
+    vmx->vmcs_pa = to;
     vmx->launched = 0;
     this_cpu(current_vmcs) = to;
 
@@ -936,11 +939,11 @@ void virtual_vmcs_enter(void *vvmcs)
 
 void virtual_vmcs_exit(void *vvmcs)
 {
-    struct vmcs_struct *cur = this_cpu(current_vmcs);
+    paddr_t cur = this_cpu(current_vmcs);
 
     __vmpclear(pfn_to_paddr(domain_page_map_to_mfn(vvmcs)));
     if ( cur )
-        __vmptrld(virt_to_maddr(cur));
+        __vmptrld(cur);
 
 }
 
@@ -1558,17 +1561,17 @@ int vmx_create_vmcs(struct vcpu *v)
     struct arch_vmx_struct *arch_vmx = &v->arch.hvm_vmx;
     int rc;
 
-    if ( (arch_vmx->vmcs = vmx_alloc_vmcs()) == NULL )
+    if ( (arch_vmx->vmcs_pa = vmx_alloc_vmcs()) == 0 )
         return -ENOMEM;
 
     INIT_LIST_HEAD(&arch_vmx->active_list);
-    __vmpclear(virt_to_maddr(arch_vmx->vmcs));
+    __vmpclear(arch_vmx->vmcs_pa);
     arch_vmx->active_cpu = -1;
     arch_vmx->launched   = 0;
 
     if ( (rc = construct_vmcs(v)) != 0 )
     {
-        vmx_free_vmcs(arch_vmx->vmcs);
+        vmx_free_vmcs(arch_vmx->vmcs_pa);
         return rc;
     }
 
@@ -1581,7 +1584,7 @@ void vmx_destroy_vmcs(struct vcpu *v)
 
     vmx_clear_vmcs(v);
 
-    vmx_free_vmcs(arch_vmx->vmcs);
+    vmx_free_vmcs(arch_vmx->vmcs_pa);
 
     free_xenheap_page(v->arch.hvm_vmx.host_msr_area);
     free_xenheap_page(v->arch.hvm_vmx.msr_area);
@@ -1612,7 +1615,7 @@ void vmx_do_resume(struct vcpu *v)
 
     if ( v->arch.hvm_vmx.active_cpu == smp_processor_id() )
     {
-        if ( v->arch.hvm_vmx.vmcs != this_cpu(current_vmcs) )
+        if ( v->arch.hvm_vmx.vmcs_pa != this_cpu(current_vmcs) )
             vmx_load_vmcs(v);
     }
     else
--- a/xen/arch/x86/hvm/vmx/vvmx.c
+++ b/xen/arch/x86/hvm/vmx/vvmx.c
@@ -56,13 +56,14 @@ int nvmx_vcpu_initialise(struct vcpu *v)
 {
     struct nestedvmx *nvmx = &vcpu_2_nvmx(v);
     struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v);
+    struct page_info *pg = alloc_domheap_page(NULL, 0);
 
-    nvcpu->nv_n2vmcx = alloc_xenheap_page();
-    if ( !nvcpu->nv_n2vmcx )
+    if ( !pg )
     {
         gdprintk(XENLOG_ERR, "nest: allocation for shadow vmcs failed\n");
         return -ENOMEM;
     }
+    nvcpu->nv_n2vmcx_pa = page_to_maddr(pg);
 
     /* non-root VMREAD/VMWRITE bitmap. */
     if ( cpu_has_vmx_vmcs_shadowing )
@@ -129,13 +130,14 @@ void nvmx_vcpu_destroy(struct vcpu *v)
      * in order to avoid double free of L2 VMCS and the possible memory
      * leak of L1 VMCS page.
      */
-    if ( nvcpu->nv_n1vmcx )
-        v->arch.hvm_vmx.vmcs = nvcpu->nv_n1vmcx;
+    if ( nvcpu->nv_n1vmcx_pa )
+        v->arch.hvm_vmx.vmcs_pa = nvcpu->nv_n1vmcx_pa;
 
-    if ( nvcpu->nv_n2vmcx ) {
-        __vmpclear(virt_to_maddr(nvcpu->nv_n2vmcx));
-        free_xenheap_page(nvcpu->nv_n2vmcx);
-        nvcpu->nv_n2vmcx = NULL;
+    if ( nvcpu->nv_n2vmcx_pa )
+    {
+        __vmpclear(nvcpu->nv_n2vmcx_pa);
+        free_domheap_page(maddr_to_page(nvcpu->nv_n2vmcx_pa));
+        nvcpu->nv_n2vmcx_pa = 0;
     }
 
     /* Must also cope with nvmx_vcpu_initialise() not having got called. */
@@ -726,8 +728,8 @@ static void __clear_current_vvmcs(struct
 {
     struct nestedvcpu *nvcpu = &vcpu_nestedhvm(v);
     
-    if ( nvcpu->nv_n2vmcx )
-        __vmpclear(virt_to_maddr(nvcpu->nv_n2vmcx));
+    if ( nvcpu->nv_n2vmcx_pa )
+        __vmpclear(nvcpu->nv_n2vmcx_pa);
 }
 
 static bool_t __must_check _map_msr_bitmap(struct vcpu *v)
@@ -1125,7 +1127,7 @@ static void virtual_vmentry(struct cpu_u
     void *vvmcs = nvcpu->nv_vvmcx;
     unsigned long lm_l1, lm_l2;
 
-    vmx_vmcs_switch(v->arch.hvm_vmx.vmcs, nvcpu->nv_n2vmcx);
+    vmx_vmcs_switch(v->arch.hvm_vmx.vmcs_pa, nvcpu->nv_n2vmcx_pa);
 
     nestedhvm_vcpu_enter_guestmode(v);
     nvcpu->nv_vmentry_pending = 0;
@@ -1335,7 +1337,7 @@ static void virtual_vmexit(struct cpu_us
          !(v->arch.hvm_vcpu.guest_efer & EFER_LMA) )
         shadow_to_vvmcs_bulk(v, ARRAY_SIZE(gpdpte_fields), gpdpte_fields);
 
-    vmx_vmcs_switch(v->arch.hvm_vmx.vmcs, nvcpu->nv_n1vmcx);
+    vmx_vmcs_switch(v->arch.hvm_vmx.vmcs_pa, nvcpu->nv_n1vmcx_pa);
 
     nestedhvm_vcpu_exit_guestmode(v);
     nvcpu->nv_vmexit_pending = 0;
@@ -1433,10 +1435,11 @@ int nvmx_handle_vmxon(struct cpu_user_re
      * `fork' the host vmcs to shadow_vmcs
      * vmcs_lock is not needed since we are on current
      */
-    nvcpu->nv_n1vmcx = v->arch.hvm_vmx.vmcs;
-    __vmpclear(virt_to_maddr(v->arch.hvm_vmx.vmcs));
-    memcpy(nvcpu->nv_n2vmcx, v->arch.hvm_vmx.vmcs, PAGE_SIZE);
-    __vmptrld(virt_to_maddr(v->arch.hvm_vmx.vmcs));
+    nvcpu->nv_n1vmcx_pa = v->arch.hvm_vmx.vmcs_pa;
+    __vmpclear(v->arch.hvm_vmx.vmcs_pa);
+    copy_domain_page(_mfn(PFN_DOWN(nvcpu->nv_n2vmcx_pa)),
+                     _mfn(PFN_DOWN(v->arch.hvm_vmx.vmcs_pa)));
+    __vmptrld(v->arch.hvm_vmx.vmcs_pa);
     v->arch.hvm_vmx.launched = 0;
     vmreturn(regs, VMSUCCEED);
 
@@ -1888,9 +1891,15 @@ int nvmx_msr_read_intercept(unsigned int
      */
     switch (msr) {
     case MSR_IA32_VMX_BASIC:
+    {
+        const struct vmcs_struct *vmcs =
+            map_domain_page(_mfn(PFN_DOWN(v->arch.hvm_vmx.vmcs_pa)));
+
         data = (host_data & (~0ul << 32)) |
-               (v->arch.hvm_vmx.vmcs->vmcs_revision_id & 0x7fffffff);
+               (vmcs->vmcs_revision_id & 0x7fffffff);
+        unmap_domain_page(vmcs);
         break;
+    }
     case MSR_IA32_VMX_PINBASED_CTLS:
     case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
         /* 1-seetings */
--- a/xen/include/asm-x86/hvm/vcpu.h
+++ b/xen/include/asm-x86/hvm/vcpu.h
@@ -104,8 +104,8 @@ struct nestedvcpu {
     void *nv_n2vmcx; /* shadow VMCB/VMCS used to run l2 guest */
 
     uint64_t nv_vvmcxaddr; /* l1 guest physical address of nv_vvmcx */
-    uint64_t nv_n1vmcx_pa; /* host physical address of nv_n1vmcx */
-    uint64_t nv_n2vmcx_pa; /* host physical address of nv_n2vmcx */
+    paddr_t nv_n1vmcx_pa; /* host physical address of nv_n1vmcx */
+    paddr_t nv_n2vmcx_pa; /* host physical address of nv_n2vmcx */
 
     /* SVM/VMX arch specific */
     union {
--- a/xen/include/asm-x86/hvm/vmx/vmcs.h
+++ b/xen/include/asm-x86/hvm/vmx/vmcs.h
@@ -92,8 +92,8 @@ struct pi_desc {
 #define NR_PML_ENTRIES   512
 
 struct arch_vmx_struct {
-    /* Virtual address of VMCS. */
-    struct vmcs_struct  *vmcs;
+    /* Physical address of VMCS. */
+    paddr_t              vmcs_pa;
     /* VMCS shadow machine address. */
     paddr_t             vmcs_shadow_maddr;
 
@@ -488,7 +488,7 @@ void vmx_enable_intercept_for_msr(struct
 int vmx_read_guest_msr(u32 msr, u64 *val);
 int vmx_write_guest_msr(u32 msr, u64 val);
 int vmx_add_msr(u32 msr, int type);
-void vmx_vmcs_switch(struct vmcs_struct *from, struct vmcs_struct *to);
+void vmx_vmcs_switch(paddr_t from, paddr_t to);
 void vmx_set_eoi_exit_bitmap(struct vcpu *v, u8 vector);
 void vmx_clear_eoi_exit_bitmap(struct vcpu *v, u8 vector);
 int vmx_check_msr_bitmap(unsigned long *msr_bitmap, u32 msr, int access_type);

Attachment: VMX-VMCS-domheap.patch
Description: Text document

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