Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Yan Zhao <yan.y.zhao@intel.com>
To:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org
Cc:     pbonzini@redhat.com, seanjc@google.com, chao.gao@intel.com,
        kai.huang@intel.com, robert.hoo.linux@gmail.com,
        Yan Zhao <yan.y.zhao@intel.com>
Subject: [PATCH v3 11/11] KVM: x86/mmu: split a single gfn zap range when guest MTRRs are honored
Date:   Fri, 16 Jun 2023 10:42:13 +0800
Message-Id: <20230616024213.7706-1-yan.y.zhao@intel.com>
In-Reply-To: <20230616023101.7019-1-yan.y.zhao@intel.com>
References: <20230616023101.7019-1-yan.y.zhao@intel.com>
Precedence: bulk

Split a single gfn zap range (specifially range [0, ~0UL)) to smaller
ranges according to current memslot layout when guest MTRRs are honored.

Though vCPUs have been serialized to perform kvm_zap_gfn_range() for MTRRs
updates and CR0.CD toggles, contention caused rescheduling cost is still
huge when there're concurrent page fault caused read locks of
kvm->mmu_lock.

Split a single huge zap range according to the actual memslot layout can
reduce unnecessary transversal and scheduling cost in tdp mmu.
Also, it can increase the chances for larger ranges to find existing ranges
to zap in zap list.

Signed-off-by: Yan Zhao <yan.y.zhao@intel.com>
---
 arch/x86/kvm/mtrr.c | 31 +++++++++++++++++++++++++------
 1 file changed, 25 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kvm/mtrr.c b/arch/x86/kvm/mtrr.c
index e2a097822a62..b83abd14ccb1 100644
--- a/arch/x86/kvm/mtrr.c
+++ b/arch/x86/kvm/mtrr.c
@@ -917,21 +917,40 @@ static void kvm_zap_or_wait_mtrr_zap_list(struct kvm *kvm)
 static void kvm_mtrr_zap_gfn_range(struct kvm_vcpu *vcpu,
 				   gfn_t gfn_start, gfn_t gfn_end)
 {
+	int idx = srcu_read_lock(&vcpu->kvm->srcu);
+	const struct kvm_memory_slot *memslot;
 	struct mtrr_zap_range *range;
+	struct kvm_memslot_iter iter;
+	struct kvm_memslots *slots;
+	gfn_t start, end;
+	int i;
+
+	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
+		slots = __kvm_memslots(vcpu->kvm, i);
+		kvm_for_each_memslot_in_gfn_range(&iter, slots, gfn_start, gfn_end) {
+			memslot = iter.slot;
+			start = max(gfn_start, memslot->base_gfn);
+			end = min(gfn_end, memslot->base_gfn + memslot->npages);
+			if (WARN_ON_ONCE(start >= end))
+				continue;
 
-	range = kmalloc(sizeof(*range), GFP_KERNEL_ACCOUNT);
-	if (!range)
-		goto fail;
+			range = kmalloc(sizeof(*range), GFP_KERNEL_ACCOUNT);
+			if (!range)
+				goto fail;
 
-	range->start = gfn_start;
-	range->end = gfn_end;
+			range->start = start;
+			range->end = end;
 
-	kvm_add_mtrr_zap_list(vcpu->kvm, range);
+			kvm_add_mtrr_zap_list(vcpu->kvm, range);
+		}
+	}
+	srcu_read_unlock(&vcpu->kvm->srcu, idx);
 
 	kvm_zap_or_wait_mtrr_zap_list(vcpu->kvm);
 	return;
 
 fail:
+	srcu_read_unlock(&vcpu->kvm->srcu, idx);
 	kvm_clear_mtrr_zap_list(vcpu->kvm);
 	kvm_zap_gfn_range(vcpu->kvm, gfn_start, gfn_end);
 }
-- 
2.17.1