Skip to contentSkip to conversation

KVM: PPC: Book3S HV P9: Optimise loads around context switch

PATCH Applied Nicholas Piggin 3 participants 3 messages Jan 23, 2022 – May 24, 2022

Subsystems (2)
lore
Share & tools
lore

Nicholas Piggin

author reviewer
#0 lore raw

PATCH 1/1 KVM: PPC: Book3S HV P9: Optimise loads around context switch

It is better to get all loads for the register values in flight before starting to switch LPID, PID, and LPCR because those mtSPRs are expensive and serialising.

This also just tidies up the code for a potential future change to the context switching sequence.

---
 arch/powerpc/kvm/book3s_hv_p9_entry.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)
arch/powerpc/kvm/book3s_hv_p9_entry.c +11 -4
index a28e5b3daabd..9dba3e3f65a0 100644
@@ -539,8 +539,10 @@ static void switch_mmu_to_guest_radix(struct kvm *kvm, struct kvm_vcpu *vcpu, u6
 {
 	struct kvm_nested_guest *nested = vcpu->arch.nested;
 	u32 lpid;
+	u32 pid;
 
 	lpid = nested ? nested->shadow_lpid : kvm->arch.lpid;
+	pid = vcpu->arch.pid;
 
 	/*
 	 * Prior memory accesses to host PID Q3 must be completed before we
@@ -551,7 +553,7 @@ static void switch_mmu_to_guest_radix(struct kvm *kvm, struct kvm_vcpu *vcpu, u6
 	isync();
 	mtspr(SPRN_LPID, lpid);
 	mtspr(SPRN_LPCR, lpcr);
-	mtspr(SPRN_PID, vcpu->arch.pid);
+	mtspr(SPRN_PID, pid);
 	/*
 	 * isync not required here because we are HRFID'ing to guest before
 	 * any guest context access, which is context synchronising.
@@ -561,9 +563,11 @@ static void switch_mmu_to_guest_radix(struct kvm *kvm, struct kvm_vcpu *vcpu, u6
 static void switch_mmu_to_guest_hpt(struct kvm *kvm, struct kvm_vcpu *vcpu, u64 lpcr)
 {
 	u32 lpid;
+	u32 pid;
 	int i;
 
 	lpid = kvm->arch.lpid;
+	pid = vcpu->arch.pid;
 
 	/*
 	 * See switch_mmu_to_guest_radix. ptesync should not be required here
@@ -574,7 +578,7 @@ static void switch_mmu_to_guest_hpt(struct kvm *kvm, struct kvm_vcpu *vcpu, u64
 	isync();
 	mtspr(SPRN_LPID, lpid);
 	mtspr(SPRN_LPCR, lpcr);
-	mtspr(SPRN_PID, vcpu->arch.pid);
+	mtspr(SPRN_PID, pid);
 
 	for (i = 0; i < vcpu->arch.slb_max; i++)
 		mtslb(vcpu->arch.slb[i].orige, vcpu->arch.slb[i].origv);
@@ -585,6 +589,9 @@ static void switch_mmu_to_guest_hpt(struct kvm *kvm, struct kvm_vcpu *vcpu, u64
 
 static void switch_mmu_to_host(struct kvm *kvm, u32 pid)
 {
+	u32 lpid = kvm->arch.host_lpid;
+	u64 lpcr = kvm->arch.host_lpcr;
+
 	/*
 	 * The guest has exited, so guest MMU context is no longer being
 	 * non-speculatively accessed, but a hwsync is needed before the
@@ -594,8 +601,8 @@ static void switch_mmu_to_host(struct kvm *kvm, u32 pid)
 	asm volatile("hwsync" ::: "memory");
 	isync();
 	mtspr(SPRN_PID, pid);
-	mtspr(SPRN_LPID, kvm->arch.host_lpid);
-	mtspr(SPRN_LPCR, kvm->arch.host_lpcr);
+	mtspr(SPRN_LPID, lpid);
+	mtspr(SPRN_LPCR, lpcr);
 	/*
 	 * isync is not required after the switch, because mtmsrd with L=0
 	 * is performed after this switch, which is context synchronising.
Signature · 1 line
2.23.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 8 lines · Nicholas Piggin

It is better to get all loads for the register values in flight before starting to switch LPID, PID, and LPCR because those mtSPRs are expensive and serialising.

This also just tidies up the code for a potential future change to the context switching sequence.

#2 lore raw
In reply to Nicholas Piggin
Quoted 8 lines · Nicholas Piggin

It is better to get all loads for the register values in flight before starting to switch LPID, PID, and LPCR because those mtSPRs are expensive and serialising.

This also just tidies up the code for a potential future change to the context switching sequence.

[...]

Applied to powerpc/topic/ppc-kvm.

[1/1] KVM: PPC: Book3S HV P9: Optimise loads around context switch

      https://git.kernel.org/powerpc/c/361234d7a1c9a5290d33e35d49821b7a32a32854

cheers