[PATCH v3 10/11] KVM: x86/mmu: fine-grained gfn zap when guest MTRRs are honored

[Yan Zhao <yan.y.zhao@intel.com>]

Find out guest MTRR ranges of non-default type and only zap those ranges
when guest MTRRs are enabled and MTRR default type equals to the memtype of
CR0.CD=1.

This allows fine-grained and faster gfn zap because of precise and shorter
zap ranges, and increases chances for concurent vCPUs to find existing
ranges to zap in zap list.

Incidentally, fix a typo in the original comment.

Suggested-by: Sean Christopherson <seanjc@google.com>
Co-developed-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Yan Zhao <yan.y.zhao@intel.com>
---
 arch/x86/kvm/mtrr.c | 108 +++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 106 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/mtrr.c b/arch/x86/kvm/mtrr.c
index 688748e3a4d2..e2a097822a62 100644
--- a/arch/x86/kvm/mtrr.c
+++ b/arch/x86/kvm/mtrr.c
@@ -179,7 +179,7 @@ static struct fixed_mtrr_segment fixed_seg_table[] = {
 	{
 		.start = 0xc0000,
 		.end = 0x100000,
-		.range_shift = 12, /* 12K */
+		.range_shift = 12, /* 4K */
 		.range_start = 24,
 	}
 };
@@ -816,6 +816,67 @@ static void kvm_add_mtrr_zap_list(struct kvm *kvm, struct mtrr_zap_range *range)
 	spin_unlock(&kvm->arch.mtrr_zap_list_lock);
 }
 
+/*
+ * Fixed ranges are only 256 pages in total.
+ * After balancing between reducing overhead of zap multiple ranges
+ * and increasing chances of finding duplicated ranges,
+ * just add fixed mtrr ranges as a whole to the mtrr zap list
+ * if memory type of one of them is not the specified type.
+ */
+static int prepare_zaplist_fixed_mtrr_of_non_type(struct kvm_vcpu *vcpu, u8 type)
+{
+	struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
+	struct mtrr_zap_range *range;
+	int index, seg_end;
+	u8 mem_type;
+
+	for (index = 0; index < KVM_NR_FIXED_MTRR_REGION; index++) {
+		mem_type = mtrr_state->fixed_ranges[index];
+
+		if (mem_type == type)
+			continue;
+
+		range = kmalloc(sizeof(*range), GFP_KERNEL_ACCOUNT);
+		if (!range)
+			return -ENOMEM;
+
+		seg_end = ARRAY_SIZE(fixed_seg_table) - 1;
+		range->start = gpa_to_gfn(fixed_seg_table[0].start);
+		range->end = gpa_to_gfn(fixed_seg_table[seg_end].end);
+		kvm_add_mtrr_zap_list(vcpu->kvm, range);
+		break;
+	}
+	return 0;
+}
+
+/*
+ * Add var mtrr ranges to the mtrr zap list
+ * if its memory type does not equal to type
+ */
+static int prepare_zaplist_var_mtrr_of_non_type(struct kvm_vcpu *vcpu, u8 type)
+{
+	struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
+	struct mtrr_zap_range *range;
+	struct kvm_mtrr_range *tmp;
+	u8 mem_type;
+
+	list_for_each_entry(tmp, &mtrr_state->head, node) {
+		mem_type = tmp->base & 0xff;
+		if (mem_type == type)
+			continue;
+
+		range = kmalloc(sizeof(*range), GFP_KERNEL_ACCOUNT);
+		if (!range)
+			return -ENOMEM;
+
+		var_mtrr_range(tmp, &range->start, &range->end);
+		range->start = gpa_to_gfn(range->start);
+		range->end = gpa_to_gfn(range->end);
+		kvm_add_mtrr_zap_list(vcpu->kvm, range);
+	}
+	return 0;
+}
+
 static void kvm_zap_mtrr_zap_list(struct kvm *kvm)
 {
 	struct list_head *head = &kvm->arch.mtrr_zap_list;
@@ -875,7 +936,50 @@ static void kvm_mtrr_zap_gfn_range(struct kvm_vcpu *vcpu,
 	kvm_zap_gfn_range(vcpu->kvm, gfn_start, gfn_end);
 }
 
+/*
+ * Zap GFN ranges when CR0.CD toggles between 0 and 1.
+ * With noncoherent DMA present,
+ * when CR0.CD=1, TDP memtype is WB or UC + IPAT;
+ * when CR0.CD=0, TDP memtype is determined by guest MTRR.
+ * Therefore, if the cache disabled memtype is different from default memtype
+ * in guest MTRR, everything is zapped;
+ * if the cache disabled memtype is equal to default memtype in guest MTRR,
+ * only MTRR ranges of non-default-memtype are required to be zapped.
+ */
 void kvm_zap_gfn_range_on_cd_toggle(struct kvm_vcpu *vcpu)
 {
-	return kvm_mtrr_zap_gfn_range(vcpu, gpa_to_gfn(0), gpa_to_gfn(~0ULL));
+	struct kvm_mtrr *mtrr_state = &vcpu->arch.mtrr_state;
+	bool mtrr_enabled = mtrr_is_enabled(mtrr_state);
+	u8 default_type;
+	u8 cd_type;
+	bool ipat;
+
+	kvm_mtrr_get_cd_memory_type(vcpu, &cd_type, &ipat);
+
+	default_type = mtrr_enabled ? mtrr_default_type(mtrr_state) :
+		       mtrr_disabled_type(vcpu);
+
+	if (cd_type != default_type || ipat)
+		return kvm_mtrr_zap_gfn_range(vcpu, gpa_to_gfn(0), gpa_to_gfn(~0ULL));
+
+	/*
+	 * If mtrr is not enabled, it will go to zap all above if the default
+	 * type does not equal to cd_type;
+	 * Or it has no need to zap if the default type equals to cd_type.
+	 */
+	if (mtrr_enabled) {
+		if (prepare_zaplist_fixed_mtrr_of_non_type(vcpu, default_type))
+			goto fail;
+
+		if (prepare_zaplist_var_mtrr_of_non_type(vcpu, default_type))
+			goto fail;
+
+		kvm_zap_or_wait_mtrr_zap_list(vcpu->kvm);
+	}
+	return;
+fail:
+	kvm_clear_mtrr_zap_list(vcpu->kvm);
+	/* resort to zapping all on failure*/
+	kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
+	return;
 }
-- 
2.17.1