[PATCH 1/2] KVM: KVM-on-hyperv: shorten no-entry section on reenlightenment

From: Paolo Bonzini
Date: Wed Aug 11 2021 - 06:24:11 EST


During re-enlightenment, update kvmclock a VM at a time instead of
raising KVM_REQ_MASTERCLOCK_UPDATE for all VMs. Because the guests
can now run after TSC emulation has been disabled, invalidate
their TSC page so that they refer to the reference time counter
MSR while the update is in progress.

Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
---
arch/x86/kvm/x86.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index bab8eb3e0a47..284afaa1db86 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -8111,7 +8111,7 @@ static void kvm_hyperv_tsc_notifier(void)

mutex_lock(&kvm_lock);
list_for_each_entry(kvm, &vm_list, vm_list)
- kvm_make_mclock_inprogress_request(kvm);
+ kvm_hv_invalidate_tsc_page(kvm);

hyperv_stop_tsc_emulation();

@@ -8123,6 +8123,7 @@ static void kvm_hyperv_tsc_notifier(void)
list_for_each_entry(kvm, &vm_list, vm_list) {
struct kvm_arch *ka = &kvm->arch;

+ kvm_make_mclock_inprogress_request(kvm);
spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
pvclock_update_vm_gtod_copy(kvm);
spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
--
2.27.0