pve-kernel-lowlatency-qoup/patches/kernel/0027-KVM-SVM-fix-tsc-scaling-cache-logic.patch

From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Maxim Levitsky <mlevitsk@redhat.com>
Date: Mon, 6 Jun 2022 21:11:49 +0300
Subject: [PATCH] KVM: SVM: fix tsc scaling cache logic

SVM uses a per-cpu variable to cache the current value of the
tsc scaling multiplier msr on each cpu.

Commit 1ab9287add5e2
("KVM: X86: Add vendor callbacks for writing the TSC multiplier")
broke this caching logic.

Refactor the code so that all TSC scaling multiplier writes go through
a single function which checks and updates the cache.

This fixes the following scenario:

1. A CPU runs a guest with some tsc scaling ratio.

2. New guest with different tsc scaling ratio starts on this CPU
   and terminates almost immediately.

   This ensures that the short running guest had set the tsc scaling ratio just
   once when it was set via KVM_SET_TSC_KHZ. Due to the bug,
   the per-cpu cache is not updated.

3. The original guest continues to run, it doesn't restore the msr
   value back to its own value, because the cache matches,
   and thus continues to run with a wrong tsc scaling ratio.

Fixes: 1ab9287add5e2 ("KVM: X86: Add vendor callbacks for writing the TSC multiplier")
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20220606181149.103072-1-mlevitsk@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
[FE: backport, mainly dropped parts for the not yet present
     5228eb96a487 ("KVM: x86: nSVM: implement nested TSC scaling")]
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
---
 arch/x86/kvm/svm/svm.c | 30 +++++++++++++++++++-----------
 1 file changed, 19 insertions(+), 11 deletions(-)

diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index cbfd9b23c32b..14d6cad2afdc 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -472,11 +472,24 @@ static int has_svm(void)
 	return 1;
 }
 
+void __svm_write_tsc_multiplier(u64 multiplier)
+{
+	preempt_disable();
+
+	if (multiplier == __this_cpu_read(current_tsc_ratio))
+		goto out;
+
+	wrmsrl(MSR_AMD64_TSC_RATIO, multiplier);
+	__this_cpu_write(current_tsc_ratio, multiplier);
+out:
+	preempt_enable();
+}
+
 static void svm_hardware_disable(void)
 {
 	/* Make sure we clean up behind us */
 	if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
-		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
+		__svm_write_tsc_multiplier(TSC_RATIO_DEFAULT);
 
 	cpu_svm_disable();
 
@@ -518,8 +531,7 @@ static int svm_hardware_enable(void)
 	wrmsrl(MSR_VM_HSAVE_PA, __sme_page_pa(sd->save_area));
 
 	if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
-		wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
-		__this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
+		__svm_write_tsc_multiplier(TSC_RATIO_DEFAULT);
 	}
 
 
@@ -1132,9 +1144,10 @@ static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
 
 static void svm_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
 {
-	wrmsrl(MSR_AMD64_TSC_RATIO, multiplier);
+	__svm_write_tsc_multiplier(multiplier);
 }
 
+
 /* Evaluate instruction intercepts that depend on guest CPUID features. */
 static void svm_recalc_instruction_intercepts(struct kvm_vcpu *vcpu,
 					      struct vcpu_svm *svm)
@@ -1457,13 +1470,8 @@ static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
 		vmsave(__sme_page_pa(sd->save_area));
 	}
 
-	if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
-		u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
-		if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
-			__this_cpu_write(current_tsc_ratio, tsc_ratio);
-			wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
-		}
-	}
+	if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
+		__svm_write_tsc_multiplier(vcpu->arch.tsc_scaling_ratio);
 
 	if (likely(tsc_aux_uret_slot >= 0))
 		kvm_set_user_return_msr(tsc_aux_uret_slot, svm->tsc_aux, -1ull);