KVM: Use alignment properties of vcpu to simplify FPU ops

Now we use a kmem cache for allocating vcpus, we can get the 16-byte
alignment required by fxsave & fxrstor instructions, and avoid
manually aligning the buffer.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Avi Kivity <avi@qumranet.com>

3 files changed, 25 insertions, 41 deletions

diff --git a/drivers/kvm/kvm.h b/drivers/kvm/kvm.h
index b362e8f8f832..7a34706f42be 100644
--- a/drivers/kvm/kvm.h
+++ b/drivers/kvm/kvm.h
@@ -45,10 +45,6 @@
 #define KVM_REFILL_PAGES 25
 #define KVM_MAX_CPUID_ENTRIES 40
 
-#define FX_IMAGE_SIZE 512
-#define FX_IMAGE_ALIGN 16
-#define FX_BUF_SIZE (2 * FX_IMAGE_SIZE + FX_IMAGE_ALIGN)
-
 #define DE_VECTOR 0
 #define NM_VECTOR 7
 #define DF_VECTOR 8
@@ -342,9 +338,8 @@ struct kvm_vcpu {
 
         struct kvm_guest_debug guest_debug;
 
-        char fx_buf[FX_BUF_SIZE];
-        char *host_fx_image;
-        char *guest_fx_image;
+        struct i387_fxsave_struct host_fx_image;
+        struct i387_fxsave_struct guest_fx_image;
         int fpu_active;
         int guest_fpu_loaded;
 
@@ -704,12 +699,12 @@ static inline unsigned long read_msr(unsigned long msr)
 }
 #endif
 
-static inline void fx_save(void *image)
+static inline void fx_save(struct i387_fxsave_struct *image)
 {
         asm ("fxsave (%0)":: "r" (image));
 }
 
-static inline void fx_restore(void *image)
+static inline void fx_restore(struct i387_fxsave_struct *image)
 {
         asm ("fxrstor (%0)":: "r" (image));
 }
diff --git a/drivers/kvm/kvm_main.c b/drivers/kvm/kvm_main.c
index 4166a08ce500..bfb1b6de0584 100644
--- a/drivers/kvm/kvm_main.c
+++ b/drivers/kvm/kvm_main.c
@@ -154,8 +154,8 @@ void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
                 return;
 
         vcpu->guest_fpu_loaded = 1;
-        fx_save(vcpu->host_fx_image);
-        fx_restore(vcpu->guest_fx_image);
+        fx_save(&vcpu->host_fx_image);
+        fx_restore(&vcpu->guest_fx_image);
 }
 EXPORT_SYMBOL_GPL(kvm_load_guest_fpu);
 
@@ -165,8 +165,8 @@ void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
                 return;
 
         vcpu->guest_fpu_loaded = 0;
-        fx_save(vcpu->guest_fx_image);
-        fx_restore(vcpu->host_fx_image);
+        fx_save(&vcpu->guest_fx_image);
+        fx_restore(&vcpu->host_fx_image);
 }
 EXPORT_SYMBOL_GPL(kvm_put_guest_fpu);
 
@@ -262,10 +262,6 @@ int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
         }
         vcpu->pio_data = page_address(page);
 
-        vcpu->host_fx_image = (char*)ALIGN((hva_t)vcpu->fx_buf,
-                                           FX_IMAGE_ALIGN);
-        vcpu->guest_fx_image = vcpu->host_fx_image + FX_IMAGE_SIZE;
-
         r = kvm_mmu_create(vcpu);
         if (r < 0)
                 goto fail_free_pio_data;
@@ -615,30 +611,20 @@ EXPORT_SYMBOL_GPL(set_cr8);
 
 void fx_init(struct kvm_vcpu *vcpu)
 {
-        struct __attribute__ ((__packed__)) fx_image_s {
-                u16 control; //fcw
-                u16 status; //fsw
-                u16 tag; // ftw
-                u16 opcode; //fop
-                u64 ip; // fpu ip
-                u64 operand;// fpu dp
-                u32 mxcsr;
-                u32 mxcsr_mask;
-
-        } *fx_image;
+        unsigned after_mxcsr_mask;
 
         /* Initialize guest FPU by resetting ours and saving into guest's */
         preempt_disable();
-        fx_save(vcpu->host_fx_image);
+        fx_save(&vcpu->host_fx_image);
         fpu_init();
-        fx_save(vcpu->guest_fx_image);
-        fx_restore(vcpu->host_fx_image);
+        fx_save(&vcpu->guest_fx_image);
+        fx_restore(&vcpu->host_fx_image);
         preempt_enable();
 
-        fx_image = (struct fx_image_s *)vcpu->guest_fx_image;
-        fx_image->mxcsr = 0x1f80;
-        memset(vcpu->guest_fx_image + sizeof(struct fx_image_s),
-               0, FX_IMAGE_SIZE - sizeof(struct fx_image_s));
+        after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
+        vcpu->guest_fx_image.mxcsr = 0x1f80;
+        memset((void *)&vcpu->guest_fx_image + after_mxcsr_mask,
+               0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 }
 EXPORT_SYMBOL_GPL(fx_init);
 
@@ -2356,6 +2342,9 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
 
         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 
+        /* We do fxsave: this must be aligned. */
+        BUG_ON((unsigned long)&vcpu->host_fx_image & 0xF);
+
         vcpu_load(vcpu);
         r = kvm_mmu_setup(vcpu);
         vcpu_put(vcpu);
@@ -2468,7 +2457,7 @@ struct fxsave {
 
 static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
-        struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
+        struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
 
         vcpu_load(vcpu);
 
@@ -2488,7 +2477,7 @@ static int kvm_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 
 static int kvm_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
-        struct fxsave *fxsave = (struct fxsave *)vcpu->guest_fx_image;
+        struct fxsave *fxsave = (struct fxsave *)&vcpu->guest_fx_image;
 
         vcpu_load(vcpu);
 
diff --git a/drivers/kvm/svm.c b/drivers/kvm/svm.c
index 8193651dd815..5277084f3a35 100644
--- a/drivers/kvm/svm.c
+++ b/drivers/kvm/svm.c
@@ -1557,8 +1557,8 @@ again:
         }
 
         if (vcpu->fpu_active) {
-                fx_save(vcpu->host_fx_image);
-                fx_restore(vcpu->guest_fx_image);
+                fx_save(&vcpu->host_fx_image);
+                fx_restore(&vcpu->guest_fx_image);
         }
 
         asm volatile (
@@ -1670,8 +1670,8 @@ again:
         vcpu->guest_mode = 0;
 
         if (vcpu->fpu_active) {
-                fx_save(vcpu->guest_fx_image);
-                fx_restore(vcpu->host_fx_image);
+                fx_save(&vcpu->guest_fx_image);
+                fx_restore(&vcpu->host_fx_image);
         }
 
         if ((svm->vmcb->save.dr7 & 0xff))