diff mbox series

KVM: PPC: Book3S HV: Don't use streamlined entry path on early POWER9 chips

Message ID 20181019100617.GA6230@blackberry
State Accepted
Headers show
Series KVM: PPC: Book3S HV: Don't use streamlined entry path on early POWER9 chips | expand

Commit Message

Paul Mackerras Oct. 19, 2018, 10:06 a.m. UTC
This disables the use of the streamlined entry path for radix guests
on early POWER9 chips that need the workaround added in commit
a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM",
2017-07-24), because the streamlined entry path does not include
that workaround.  This also means that we can't do nested HV-KVM
on those chips.

Since the chips that need that workaround are the same ones that can't
run both radix and HPT guests at the same time on different threads of
a core, we use the existing 'no_mixing_hpt_and_radix' variable that
identifies those chips to identify when we can't use the new guest
entry path, and when we can't do nested virtualization.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
---
 arch/powerpc/kvm/book3s_hv.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)
diff mbox series

Patch

diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index 788bc61..bf8def2 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -4174,7 +4174,16 @@  static int kvmppc_vcpu_run_hv(struct kvm_run *run, struct kvm_vcpu *vcpu)
 	vcpu->arch.state = KVMPPC_VCPU_BUSY_IN_HOST;
 
 	do {
-		if (kvm->arch.threads_indep && kvm_is_radix(kvm))
+		/*
+		 * The early POWER9 chips that can't mix radix and HPT threads
+		 * on the same core also need the workaround for the problem
+		 * where the TLB would prefetch entries in the guest exit path
+		 * for radix guests using the guest PIDR value and LPID 0.
+		 * The workaround is in the old path (kvmppc_run_vcpu())
+		 * but not the new path (kvmhv_run_single_vcpu()).
+		 */
+		if (kvm->arch.threads_indep && kvm_is_radix(kvm) &&
+		    !no_mixing_hpt_and_radix)
 			r = kvmhv_run_single_vcpu(run, vcpu, ~(u64)0,
 						  vcpu->arch.vcore->lpcr);
 		else
@@ -5196,7 +5205,7 @@  static int kvmhv_enable_nested(struct kvm *kvm)
 {
 	if (!nested)
 		return -EPERM;
-	if (!cpu_has_feature(CPU_FTR_ARCH_300))
+	if (!cpu_has_feature(CPU_FTR_ARCH_300) || no_mixing_hpt_and_radix)
 		return -ENODEV;
 
 	/* kvm == NULL means the caller is testing if the capability exists */