[09/11] KVM: MMU: remove the mismatch shadow page

Commit Message

Xiao Guangrong July 26, 2011, 11:31 a.m. UTC

If the shadow page has different cpu mode with current vcpu, we do better
remove them since the OS does not changes cpu mode frequently

Signed-off-by: Xiao Guangrong <[e-mail]>
---
 arch/x86/kvm/mmu.c |   26 +++++++++++++++++++-------
 1 files changed, 19 insertions(+), 7 deletions(-)

Comments

Avi Kivity July 27, 2011, 9:11 a.m. UTC | #1

On 07/26/2011 02:31 PM, Xiao Guangrong wrote:
> If the shadow page has different cpu mode with current vcpu, we do better
> remove them since the OS does not changes cpu mode frequently
>

This is true, but is it really needed?  Won't such pages be torn down 
very quickly anyway, and will never be reestablished?

Avi Kivity July 27, 2011, 9:13 a.m. UTC | #2

On 07/27/2011 12:11 PM, Avi Kivity wrote:
> On 07/26/2011 02:31 PM, Xiao Guangrong wrote:
>> If the shadow page has different cpu mode with current vcpu, we do 
>> better
>> remove them since the OS does not changes cpu mode frequently
>>
>
> This is true

It's actually not true.  With nested virtualization the cpu changes 
modes all the time.

Xiao Guangrong July 27, 2011, 10:05 a.m. UTC | #3

On 07/27/2011 05:13 PM, Avi Kivity wrote:
> On 07/27/2011 12:11 PM, Avi Kivity wrote:
>> On 07/26/2011 02:31 PM, Xiao Guangrong wrote:
>>> If the shadow page has different cpu mode with current vcpu, we do better
>>> remove them since the OS does not changes cpu mode frequently
>>>
>>
>> This is true
> 
> It's actually not true.  With nested virtualization the cpu changes modes all the time.
> 

Oh, yes, i forgot this, thanks point it out!

--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
index 931c23a..2328ee6 100644
--- a/arch/x86/kvm/mmu.c
+++ b/arch/x86/kvm/mmu.c
@@ -3603,6 +3603,18 @@  static bool detect_write_misaligned(struct kvm_mmu_page *sp, gpa_t gpa,
 	return misaligned;
 }
 
+/*
+ * The OS hardly changes cpu mode after boot, we can zap the shadow page if
+ * it is mismatched with the current vcpu.
+ */
+static bool detect_mismatch_sp(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
+{
+	union kvm_mmu_page_role mask;
+
+	mask.cr0_wp = mask.cr4_pae = mask.nxe = 1;
+	return (sp->role.word ^ vcpu->arch.mmu.base_role.word) & mask.word;
+}
+
 static u64 *get_written_sptes(struct kvm_mmu_page *sp, gpa_t gpa, int *nspte)
 {
 	unsigned page_offset, quadrant;
@@ -3638,13 +3650,12 @@  void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 		       const u8 *new, int bytes, bool repeat_write)
 {
 	gfn_t gfn = gpa >> PAGE_SHIFT;
-	union kvm_mmu_page_role mask = { .word = 0 };
 	struct kvm_mmu_page *sp;
 	struct hlist_node *node;
 	LIST_HEAD(invalid_list);
 	u64 entry, gentry, *spte;
 	int npte;
-	bool remote_flush, local_flush, zap_page, flooded, misaligned;
+	bool remote_flush, local_flush, zap_page, flooded;
 
 	/*
 	 * If we don't have indirect shadow pages, it means no page is
@@ -3664,10 +3675,13 @@  void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 	trace_kvm_mmu_audit(vcpu, AUDIT_PRE_PTE_WRITE);
 
 	flooded = detect_write_flooding(vcpu, gfn);
-	mask.cr0_wp = mask.cr4_pae = mask.nxe = 1;
 	for_each_gfn_indirect_valid_sp(vcpu->kvm, sp, gfn, node) {
+		bool mismatch, misaligned;
+
 		misaligned = detect_write_misaligned(sp, gpa, bytes);
-		if (misaligned || flooded || repeat_write) {
+		mismatch = detect_mismatch_sp(vcpu, sp);
+
+		if (misaligned || mismatch || flooded || repeat_write) {
 			zap_page |= !!kvm_mmu_prepare_zap_page(vcpu->kvm, sp,
 						     &invalid_list);
 			++vcpu->kvm->stat.mmu_flooded;
@@ -3682,9 +3696,7 @@  void kvm_mmu_pte_write(struct kvm_vcpu *vcpu, gpa_t gpa,
 		while (npte--) {
 			entry = *spte;
 			mmu_page_zap_pte(vcpu->kvm, sp, spte);
-			if (gentry &&
-			      !((sp->role.word ^ vcpu->arch.mmu.base_role.word)
-			      & mask.word) && get_free_pte_list_desc_nr(vcpu))
+			if (gentry && get_free_pte_list_desc_nr(vcpu))
 				mmu_pte_write_new_pte(vcpu, sp, spte, &gentry);
 			if (!remote_flush && need_remote_flush(entry, *spte))
 				remote_flush = true;