[PATCH][v2 0/3] Convert KVM to use FPU API

[PATCH][v2 0/3] Convert KVM to use FPU API

[Sheng Yang]

Change from v1:
Use unlazy_fpu() to handle host FPU, avoiding save/restore of host FPU states.

Sheng Yang (3):
  x86: Export FPU API for KVM use
  KVM: x86: Use unlazy_fpu() for host FPU
  KVM: x86: Use FPU API

 arch/x86/include/asm/i387.h     |    2 +
 arch/x86/include/asm/kvm_host.h |   18 +---------
 arch/x86/include/asm/xsave.h    |    3 ++
 arch/x86/kernel/i387.c          |    3 +-
 arch/x86/kernel/process.c       |    1 +
 arch/x86/kvm/x86.c              |   70 ++++++++++-----------------------------
 6 files changed, 27 insertions(+), 70 deletions(-)

[PATCH][v2 1/3] x86: Export FPU API for KVM use

[Sheng Yang]

Also add some constants.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
---
 arch/x86/include/asm/i387.h  |    2 ++
 arch/x86/include/asm/xsave.h |    3 +++
 arch/x86/kernel/i387.c       |    3 ++-
 arch/x86/kernel/process.c    |    1 +
 4 files changed, 8 insertions(+), 1 deletions(-)

diff --git a/arch/x86/include/asm/i387.h b/arch/x86/include/asm/i387.h
index 8002e9c..9610628 100644
--- a/arch/x86/include/asm/i387.h
+++ b/arch/x86/include/asm/i387.h
@@ -488,6 +488,8 @@ static inline void fpu_copy(struct fpu *dst, struct fpu *src)
 	memcpy(dst->state, src->state, xstate_size);
 }
 
+extern void fpu_finit(struct fpu *fpu);
+
 #endif /* __ASSEMBLY__ */
 
 #define PSHUFB_XMM5_XMM0 .byte 0x66, 0x0f, 0x38, 0x00, 0xc5
diff --git a/arch/x86/include/asm/xsave.h b/arch/x86/include/asm/xsave.h
index 2c4390c..29ee4e4 100644
--- a/arch/x86/include/asm/xsave.h
+++ b/arch/x86/include/asm/xsave.h
@@ -13,6 +13,9 @@
 
 #define FXSAVE_SIZE	512
 
+#define XSTATE_YMM_SIZE 256
+#define XSTATE_YMM_OFFSET (512 + 64)
+
 /*
  * These are the features that the OS can handle currently.
  */
diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
index 86cef6b..c4444bc 100644
--- a/arch/x86/kernel/i387.c
+++ b/arch/x86/kernel/i387.c
@@ -107,7 +107,7 @@ void __cpuinit fpu_init(void)
 }
 #endif	/* CONFIG_X86_64 */
 
-static void fpu_finit(struct fpu *fpu)
+void fpu_finit(struct fpu *fpu)
 {
 #ifdef CONFIG_X86_32
 	if (!HAVE_HWFP) {
@@ -132,6 +132,7 @@ static void fpu_finit(struct fpu *fpu)
 		fp->fos = 0xffff0000u;
 	}
 }
+EXPORT_SYMBOL_GPL(fpu_finit);
 
 /*
  * The _current_ task is using the FPU for the first time
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index 8bcc21f..373fec9 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -28,6 +28,7 @@ unsigned long idle_nomwait;
 EXPORT_SYMBOL(idle_nomwait);
 
 struct kmem_cache *task_xstate_cachep;
+EXPORT_SYMBOL(task_xstate_cachep);
 
 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 {
-- 
1.7.0.1

[PATCH][v2 2/3] KVM: x86: Use unlazy_fpu() for host FPU

[Sheng Yang]

We can avoid unnecessary fpu load when userspace process
didn't use FPU frequently.

Derived from Avi's idea.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
---
 arch/x86/include/asm/kvm_host.h |    1 -
 arch/x86/kvm/x86.c              |   18 ++----------------
 2 files changed, 2 insertions(+), 17 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 0c06148..d93601c 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -301,7 +301,6 @@ struct kvm_vcpu_arch {
 		unsigned long mmu_seq;
 	} update_pte;
 
-	struct i387_fxsave_struct host_fx_image;
 	struct i387_fxsave_struct guest_fx_image;
 
 	gva_t mmio_fault_cr2;
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 7500cba..ba8c2d2 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -52,6 +52,7 @@
 #include <asm/desc.h>
 #include <asm/mtrr.h>
 #include <asm/mce.h>
+#include <asm/i387.h>
 
 #define MAX_IO_MSRS 256
 #define CR0_RESERVED_BITS						\
@@ -5134,21 +5135,10 @@ void fx_init(struct kvm_vcpu *vcpu)
 {
 	unsigned after_mxcsr_mask;
 
-	/*
-	 * Touch the fpu the first time in non atomic context as if
-	 * this is the first fpu instruction the exception handler
-	 * will fire before the instruction returns and it'll have to
-	 * allocate ram with GFP_KERNEL.
-	 */
-	if (!used_math())
-		kvm_fx_save(&vcpu->arch.host_fx_image);
-
 	/* Initialize guest FPU by resetting ours and saving into guest's */
 	preempt_disable();
-	kvm_fx_save(&vcpu->arch.host_fx_image);
 	kvm_fx_finit();
 	kvm_fx_save(&vcpu->arch.guest_fx_image);
-	kvm_fx_restore(&vcpu->arch.host_fx_image);
 	preempt_enable();
 
 	vcpu->arch.cr0 |= X86_CR0_ET;
@@ -5165,7 +5155,7 @@ void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 		return;
 
 	vcpu->guest_fpu_loaded = 1;
-	kvm_fx_save(&vcpu->arch.host_fx_image);
+	unlazy_fpu(current);
 	kvm_fx_restore(&vcpu->arch.guest_fx_image);
 	trace_kvm_fpu(1);
 }
@@ -5177,7 +5167,6 @@ void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 
 	vcpu->guest_fpu_loaded = 0;
 	kvm_fx_save(&vcpu->arch.guest_fx_image);
-	kvm_fx_restore(&vcpu->arch.host_fx_image);
 	++vcpu->stat.fpu_reload;
 	set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
 	trace_kvm_fpu(0);
@@ -5203,9 +5192,6 @@ int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
 {
 	int r;
 
-	/* We do fxsave: this must be aligned. */
-	BUG_ON((unsigned long)&vcpu->arch.host_fx_image & 0xF);
-
 	vcpu->arch.mtrr_state.have_fixed = 1;
 	vcpu_load(vcpu);
 	r = kvm_arch_vcpu_reset(vcpu);
-- 
1.7.0.1

[PATCH][v2 3/3] KVM: x86: Use FPU API

[Sheng Yang]

Convert KVM to use generic FPU API.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
---
 arch/x86/include/asm/kvm_host.h |   17 +------------
 arch/x86/kvm/x86.c              |   52 ++++++++++++---------------------------
 2 files changed, 17 insertions(+), 52 deletions(-)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index d93601c..d08bb4a 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -301,7 +301,7 @@ struct kvm_vcpu_arch {
 		unsigned long mmu_seq;
 	} update_pte;
 
-	struct i387_fxsave_struct guest_fx_image;
+	struct fpu guest_fpu;
 
 	gva_t mmio_fault_cr2;
 	struct kvm_pio_request pio;
@@ -708,21 +708,6 @@ static inline unsigned long read_msr(unsigned long msr)
 }
 #endif
 
-static inline void kvm_fx_save(struct i387_fxsave_struct *image)
-{
-	asm("fxsave (%0)":: "r" (image));
-}
-
-static inline void kvm_fx_restore(struct i387_fxsave_struct *image)
-{
-	asm("fxrstor (%0)":: "r" (image));
-}
-
-static inline void kvm_fx_finit(void)
-{
-	asm("finit");
-}
-
 static inline u32 get_rdx_init_val(void)
 {
 	return 0x600; /* P6 family */
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index ba8c2d2..7be1d36 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -53,6 +53,7 @@
 #include <asm/mtrr.h>
 #include <asm/mce.h>
 #include <asm/i387.h>
+#include <asm/xcr.h>
 
 #define MAX_IO_MSRS 256
 #define CR0_RESERVED_BITS						\
@@ -5058,27 +5059,6 @@ out:
 }
 
 /*
- * fxsave fpu state.  Taken from x86_64/processor.h.  To be killed when
- * we have asm/x86/processor.h
- */
-struct fxsave {
-	u16	cwd;
-	u16	swd;
-	u16	twd;
-	u16	fop;
-	u64	rip;
-	u64	rdp;
-	u32	mxcsr;
-	u32	mxcsr_mask;
-	u32	st_space[32];	/* 8*16 bytes for each FP-reg = 128 bytes */
-#ifdef CONFIG_X86_64
-	u32	xmm_space[64];	/* 16*16 bytes for each XMM-reg = 256 bytes */
-#else
-	u32	xmm_space[32];	/* 8*16 bytes for each XMM-reg = 128 bytes */
-#endif
-};
-
-/*
  * Translate a guest virtual address to a guest physical address.
  */
 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
@@ -5101,7 +5081,8 @@ int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 
 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
-	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
+	struct i387_fxsave_struct *fxsave =
+			&vcpu->arch.guest_fpu.state->fxsave;
 
 	memcpy(fpu->fpr, fxsave->st_space, 128);
 	fpu->fcw = fxsave->cwd;
@@ -5117,7 +5098,8 @@ int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 
 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 {
-	struct fxsave *fxsave = (struct fxsave *)&vcpu->arch.guest_fx_image;
+	struct i387_fxsave_struct *fxsave =
+			&vcpu->arch.guest_fpu.state->fxsave;
 
 	memcpy(fxsave->st_space, fpu->fpr, 128);
 	fxsave->cwd = fpu->fcw;
@@ -5133,22 +5115,18 @@ int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
 
 void fx_init(struct kvm_vcpu *vcpu)
 {
-	unsigned after_mxcsr_mask;
-
-	/* Initialize guest FPU by resetting ours and saving into guest's */
-	preempt_disable();
-	kvm_fx_finit();
-	kvm_fx_save(&vcpu->arch.guest_fx_image);
-	preempt_enable();
+	fpu_alloc(&vcpu->arch.guest_fpu);
+	fpu_finit(&vcpu->arch.guest_fpu);
 
 	vcpu->arch.cr0 |= X86_CR0_ET;
-	after_mxcsr_mask = offsetof(struct i387_fxsave_struct, st_space);
-	vcpu->arch.guest_fx_image.mxcsr = 0x1f80;
-	memset((void *)&vcpu->arch.guest_fx_image + after_mxcsr_mask,
-	       0, sizeof(struct i387_fxsave_struct) - after_mxcsr_mask);
 }
 EXPORT_SYMBOL_GPL(fx_init);
 
+static void fx_free(struct kvm_vcpu *vcpu)
+{
+	fpu_free(&vcpu->arch.guest_fpu);
+}
+
 void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 {
 	if (vcpu->guest_fpu_loaded)
@@ -5156,7 +5134,7 @@ void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
 
 	vcpu->guest_fpu_loaded = 1;
 	unlazy_fpu(current);
-	kvm_fx_restore(&vcpu->arch.guest_fx_image);
+	fpu_restore_checking(&vcpu->arch.guest_fpu);
 	trace_kvm_fpu(1);
 }
 
@@ -5166,7 +5144,7 @@ void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 		return;
 
 	vcpu->guest_fpu_loaded = 0;
-	kvm_fx_save(&vcpu->arch.guest_fx_image);
+	fpu_save_init(&vcpu->arch.guest_fpu);
 	++vcpu->stat.fpu_reload;
 	set_bit(KVM_REQ_DEACTIVATE_FPU, &vcpu->requests);
 	trace_kvm_fpu(0);
@@ -5179,6 +5157,7 @@ void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 		vcpu->arch.time_page = NULL;
 	}
 
+	fx_free(vcpu);
 	kvm_x86_ops->vcpu_free(vcpu);
 }
 
@@ -5213,6 +5192,7 @@ void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 	kvm_mmu_unload(vcpu);
 	vcpu_put(vcpu);
 
+	fx_free(vcpu);
 	kvm_x86_ops->vcpu_free(vcpu);
 }
 
-- 
1.7.0.1

Re: [PATCH][v2 1/3] x86: Export FPU API for KVM use

[Avi Kivity]

On 05/17/2010 12:08 PM, Sheng Yang wrote:
> Also add some constants.
>    

> +++ b/arch/x86/kernel/process.c
> @@ -28,6 +28,7 @@ unsigned long idle_nomwait;
>   EXPORT_SYMBOL(idle_nomwait);
>
>   struct kmem_cache *task_xstate_cachep;
> +EXPORT_SYMBOL(task_xstate_cachep);
>
>   int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
>   {
>    

_GPL() unless good reason not to.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.

Re: [PATCH][v2 1/3] x86: Export FPU API for KVM use

[Sheng Yang]

On Monday 17 May 2010 17:19:02 Avi Kivity wrote:
> On 05/17/2010 12:08 PM, Sheng Yang wrote:
> > Also add some constants.
> > 
> > 
> > +++ b/arch/x86/kernel/process.c
> > @@ -28,6 +28,7 @@ unsigned long idle_nomwait;
> > 
> >   EXPORT_SYMBOL(idle_nomwait);
> >   
> >   struct kmem_cache *task_xstate_cachep;
> > 
> > +EXPORT_SYMBOL(task_xstate_cachep);
> > 
> >   int arch_dup_task_struct(struct task_struct *dst, struct task_struct
> >   *src) {
> 
> _GPL() unless good reason not to.

Oops... Just copied and pasted the above line...

--
regards
Yang, Sheng

[PATCH][v2 1/3] x86: Export FPU API for KVM use

[Sheng Yang]

Also add some constants.

Signed-off-by: Sheng Yang <sheng@linux.intel.com>
---
 arch/x86/include/asm/i387.h  |    2 ++
 arch/x86/include/asm/xsave.h |    3 +++
 arch/x86/kernel/i387.c       |    3 ++-
 arch/x86/kernel/process.c    |    1 +
 4 files changed, 8 insertions(+), 1 deletions(-)

diff --git a/arch/x86/include/asm/i387.h b/arch/x86/include/asm/i387.h
index 8002e9c..9610628 100644
--- a/arch/x86/include/asm/i387.h
+++ b/arch/x86/include/asm/i387.h
@@ -488,6 +488,8 @@ static inline void fpu_copy(struct fpu *dst, struct fpu *src)
 	memcpy(dst->state, src->state, xstate_size);
 }
 
+extern void fpu_finit(struct fpu *fpu);
+
 #endif /* __ASSEMBLY__ */
 
 #define PSHUFB_XMM5_XMM0 .byte 0x66, 0x0f, 0x38, 0x00, 0xc5
diff --git a/arch/x86/include/asm/xsave.h b/arch/x86/include/asm/xsave.h
index 2c4390c..29ee4e4 100644
--- a/arch/x86/include/asm/xsave.h
+++ b/arch/x86/include/asm/xsave.h
@@ -13,6 +13,9 @@
 
 #define FXSAVE_SIZE	512
 
+#define XSTATE_YMM_SIZE 256
+#define XSTATE_YMM_OFFSET (512 + 64)
+
 /*
  * These are the features that the OS can handle currently.
  */
diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c
index 86cef6b..c4444bc 100644
--- a/arch/x86/kernel/i387.c
+++ b/arch/x86/kernel/i387.c
@@ -107,7 +107,7 @@ void __cpuinit fpu_init(void)
 }
 #endif	/* CONFIG_X86_64 */
 
-static void fpu_finit(struct fpu *fpu)
+void fpu_finit(struct fpu *fpu)
 {
 #ifdef CONFIG_X86_32
 	if (!HAVE_HWFP) {
@@ -132,6 +132,7 @@ static void fpu_finit(struct fpu *fpu)
 		fp->fos = 0xffff0000u;
 	}
 }
+EXPORT_SYMBOL_GPL(fpu_finit);
 
 /*
  * The _current_ task is using the FPU for the first time
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c
index 8bcc21f..0f331f4 100644
--- a/arch/x86/kernel/process.c
+++ b/arch/x86/kernel/process.c
@@ -28,6 +28,7 @@ unsigned long idle_nomwait;
 EXPORT_SYMBOL(idle_nomwait);
 
 struct kmem_cache *task_xstate_cachep;
+EXPORT_SYMBOL_GPL(task_xstate_cachep);
 
 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 {
-- 
1.7.0.1

Re: [PATCH][v2 1/3] x86: Export FPU API for KVM use

[Avi Kivity]

On 05/17/2010 12:19 PM, Sheng Yang wrote:
> On Monday 17 May 2010 17:19:02 Avi Kivity wrote:
>    
>> On 05/17/2010 12:08 PM, Sheng Yang wrote:
>>      
>>> Also add some constants.
>>>
>>>
>>> +++ b/arch/x86/kernel/process.c
>>> @@ -28,6 +28,7 @@ unsigned long idle_nomwait;
>>>
>>>    EXPORT_SYMBOL(idle_nomwait);
>>>
>>>    struct kmem_cache *task_xstate_cachep;
>>>
>>> +EXPORT_SYMBOL(task_xstate_cachep);
>>>
>>>    int arch_dup_task_struct(struct task_struct *dst, struct task_struct
>>>    *src) {
>>>        
>> _GPL() unless good reason not to.
>>      
> Oops... Just copied and pasted the above line...
>
>    

No matter, I'll update it when applying if no further comments 
(reviewing 3 now).

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.

Re: [PATCH][v2 3/3] KVM: x86: Use FPU API

[Avi Kivity]

On 05/17/2010 12:08 PM, Sheng Yang wrote:
> Convert KVM to use generic FPU API.
>
>   void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
>   {
>   	if (vcpu->guest_fpu_loaded)
> @@ -5156,7 +5134,7 @@ void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
>
>   	vcpu->guest_fpu_loaded = 1;
>   	unlazy_fpu(current);
> -	kvm_fx_restore(&vcpu->arch.guest_fx_image);
> +	fpu_restore_checking(&vcpu->arch.guest_fpu);
>   	trace_kvm_fpu(1);
>   }
>    

Do we need to do something on error here?

I think an error can only occur if userspace sets bad fpu state, so 
maybe do the check in the ioctl.

In any case, unrelated to this patch.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.

Re: [PATCH][v2 0/3] Convert KVM to use FPU API

[Avi Kivity]

On 05/17/2010 12:08 PM, Sheng Yang wrote:
> Change from v1:
> Use unlazy_fpu() to handle host FPU, avoiding save/restore of host FPU states.
>
>    

Looks good, will wait a bit for more reviews and apply.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.

Re: [PATCH][v2 0/3] Convert KVM to use FPU API

[Avi Kivity]

On 05/17/2010 12:37 PM, Avi Kivity wrote:
> On 05/17/2010 12:08 PM, Sheng Yang wrote:
>> Change from v1:
>> Use unlazy_fpu() to handle host FPU, avoiding save/restore of host 
>> FPU states.
>>
>
> Looks good, will wait a bit for more reviews and apply.
>

Now applied.  Thanks.

-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.