x86/svm: Use physical addresses for HSA and Host VMCB

They are only referenced by physical address (either the HSA MSR, or via
VMSAVE/VMLOAD which take a physical operand).  Allocating xenheap pages and
storing their virtual address is wasteful.

Allocate them with domheap pages instead, taking the opportunity to suitably
NUMA-position them.  This avoids Xen needing to perform a virt to phys
translation on every context switch.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>

diff --git a/xen/arch/x86/hvm/svm/svm.c b/xen/arch/x86/hvm/svm/svm.c
index 0dc94427e387fef32ff3967a698d123504c14bd5..599a8d32923b7bdd48e8ed4277375860e6a78f25 100644 (file)
--- a/xen/arch/x86/hvm/svm/svm.c
+++ b/xen/arch/x86/hvm/svm/svm.c
@@ -72,11 +72,13 @@ static void svm_update_guest_efer(struct vcpu *);
 
 static struct hvm_function_table svm_function_table;
 
-/* va of hardware host save area     */
-static DEFINE_PER_CPU_READ_MOSTLY(void *, hsa);
-
-/* vmcb used for extended host state */
-static DEFINE_PER_CPU_READ_MOSTLY(void *, root_vmcb);
+/*
+ * Physical addresses of the Host State Area (for hardware) and vmcb (for Xen)
+ * which contains Xen's fs/gs/tr/ldtr and GSBASE/STAR/SYSENTER state when in
+ * guest vcpu context.
+ */
+static DEFINE_PER_CPU_READ_MOSTLY(paddr_t, hsa);
+static DEFINE_PER_CPU_READ_MOSTLY(paddr_t, host_vmcb);
 
 static bool_t amd_erratum383_found __read_mostly;
 
@@ -1015,7 +1017,7 @@ static void svm_ctxt_switch_from(struct vcpu *v)
     svm_tsc_ratio_save(v);
 
     svm_sync_vmcb(v);
-    svm_vmload(per_cpu(root_vmcb, cpu));
+    svm_vmload_pa(per_cpu(host_vmcb, cpu));
 
     /* Resume use of ISTs now that the host TR is reinstated. */
     set_ist(&idt_tables[cpu][TRAP_double_fault],  IST_DF);
@@ -1045,7 +1047,7 @@ static void svm_ctxt_switch_to(struct vcpu *v)
 
     svm_restore_dr(v);
 
-    svm_vmsave(per_cpu(root_vmcb, cpu));
+    svm_vmsave_pa(per_cpu(host_vmcb, cpu));
     svm_vmload(vmcb);
     vmcb->cleanbits.bytes = 0;
     svm_lwp_load(v);
@@ -1468,24 +1470,58 @@ static int svm_event_pending(struct vcpu *v)
 
 static void svm_cpu_dead(unsigned int cpu)
 {
-    free_xenheap_page(per_cpu(hsa, cpu));
-    per_cpu(hsa, cpu) = NULL;
-    free_vmcb(per_cpu(root_vmcb, cpu));
-    per_cpu(root_vmcb, cpu) = NULL;
+    paddr_t *this_hsa = &per_cpu(hsa, cpu);
+    paddr_t *this_vmcb = &per_cpu(host_vmcb, cpu);
+
+    if ( *this_hsa )
+    {
+        free_domheap_page(maddr_to_page(*this_hsa));
+        *this_hsa = 0;
+    }
+
+    if ( *this_vmcb )
+    {
+        free_domheap_page(maddr_to_page(*this_vmcb));
+        *this_vmcb = 0;
+    }
 }
 
 static int svm_cpu_up_prepare(unsigned int cpu)
 {
-    if ( ((per_cpu(hsa, cpu) == NULL) &&
-          ((per_cpu(hsa, cpu) = alloc_host_save_area()) == NULL)) ||
-         ((per_cpu(root_vmcb, cpu) == NULL) &&
-          ((per_cpu(root_vmcb, cpu) = alloc_vmcb()) == NULL)) )
+    paddr_t *this_hsa = &per_cpu(hsa, cpu);
+    paddr_t *this_vmcb = &per_cpu(host_vmcb, cpu);
+    nodeid_t node = cpu_to_node(cpu);
+    unsigned int memflags = 0;
+    struct page_info *pg;
+
+    if ( node != NUMA_NO_NODE )
+        memflags = MEMF_node(node);
+
+    if ( !*this_hsa )
+    {
+        pg = alloc_domheap_page(NULL, memflags);
+        if ( !pg )
+            goto err;
+
+        clear_domain_page(_mfn(page_to_mfn(pg)));
+        *this_hsa = page_to_maddr(pg);
+    }
+
+    if ( !*this_vmcb )
     {
-        svm_cpu_dead(cpu);
-        return -ENOMEM;
+        pg = alloc_domheap_page(NULL, memflags);
+        if ( !pg )
+            goto err;
+
+        clear_domain_page(_mfn(page_to_mfn(pg)));
+        *this_vmcb = page_to_maddr(pg);
     }
 
     return 0;
+
+ err:
+    svm_cpu_dead(cpu);
+    return -ENOMEM;
 }
 
 static void svm_init_erratum_383(const struct cpuinfo_x86 *c)
@@ -1544,7 +1580,7 @@ static int _svm_cpu_up(bool bsp)
     write_efer(read_efer() | EFER_SVME);
 
     /* Initialize the HSA for this core. */
-    wrmsrl(MSR_K8_VM_HSAVE_PA, (uint64_t)virt_to_maddr(per_cpu(hsa, cpu)));
+    wrmsrl(MSR_K8_VM_HSAVE_PA, per_cpu(hsa, cpu));
 
     /* check for erratum 383 */
     svm_init_erratum_383(c);

diff --git a/xen/arch/x86/hvm/svm/vmcb.c b/xen/arch/x86/hvm/svm/vmcb.c
index 9493215ddf6742cde5c0eb748f2e673d139e84a3..997e7597e0c6c9b8b0f45cfc15ec2002e01cc35c 100644 (file)
--- a/xen/arch/x86/hvm/svm/vmcb.c
+++ b/xen/arch/x86/hvm/svm/vmcb.c
@@ -50,21 +50,6 @@ void free_vmcb(struct vmcb_struct *vmcb)
     free_xenheap_page(vmcb);
 }
 
-struct host_save_area *alloc_host_save_area(void)
-{
-    struct host_save_area *hsa;
-
-    hsa = alloc_xenheap_page();
-    if ( hsa == NULL )
-    {
-        printk(XENLOG_WARNING "Warning: failed to allocate hsa.\n");
-        return NULL;
-    }
-
-    clear_page(hsa);
-    return hsa;
-}
-
 /* This function can directly access fields which are covered by clean bits. */
 static int construct_vmcb(struct vcpu *v)
 {

diff --git a/xen/include/asm-x86/hvm/svm/vmcb.h b/xen/include/asm-x86/hvm/svm/vmcb.h
index ec22d9113a42d0511fa581cd6aa4535ab7f23a91..01ce20b0bdfda6d720b8a8cf674c3ac121af3df2 100644 (file)
--- a/xen/include/asm-x86/hvm/svm/vmcb.h
+++ b/xen/include/asm-x86/hvm/svm/vmcb.h
@@ -526,7 +526,6 @@ struct arch_svm_struct {
 };
 
 struct vmcb_struct *alloc_vmcb(void);
-struct host_save_area *alloc_host_save_area(void);
 void free_vmcb(struct vmcb_struct *vmcb);
 
 int  svm_create_vmcb(struct vcpu *v);