KVM: PPC: Book3S PR: Keep volatile reg values in vcpu rather than shadow_vcpu

Currently PR-style KVM keeps the volatile guest register values
(R0 - R13, CR, LR, CTR, XER, PC) in a shadow_vcpu struct rather than
the main kvm_vcpu struct.  For 64-bit, the shadow_vcpu exists in two
places, a kmalloc'd struct and in the PACA, and it gets copied back
and forth in kvmppc_core_vcpu_load/put(), because the real-mode code
can't rely on being able to access the kmalloc'd struct.

This changes the code to copy the volatile values into the shadow_vcpu
as one of the last things done before entering the guest.  Similarly
the values are copied back out of the shadow_vcpu to the kvm_vcpu
immediately after exiting the guest.  We arrange for interrupts to be
still disabled at this point so that we can't get preempted on 64-bit
and end up copying values from the wrong PACA.

This means that the accessor functions in kvm_book3s.h for these
registers are greatly simplified, and are same between PR and HV KVM.
In places where accesses to shadow_vcpu fields are now replaced by
accesses to the kvm_vcpu, we can also remove the svcpu_get/put pairs.
Finally, on 64-bit, we don't need the kmalloc'd struct at all any more.

With this, the time to read the PVR one million times in a loop went
from 567.7ms to 575.5ms (averages of 6 values), an increase of about
1.4% for this worse-case test for guest entries and exits.  The
standard deviation of the measurements is about 11ms, so the
difference is only marginally significant statistically.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>

5 files changed, 121 insertions, 49 deletions

diff --git a/arch/powerpc/kvm/book3s_emulate.c b/arch/powerpc/kvm/book3s_emulate.c
index 360ce68c9809..34044b111daa 100644
--- a/arch/powerpc/kvm/book3s_emulate.c
+++ b/arch/powerpc/kvm/book3s_emulate.c
@@ -267,12 +267,9 @@ int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
 
 			r = kvmppc_st(vcpu, &addr, 32, zeros, true);
 			if ((r == -ENOENT) || (r == -EPERM)) {
-				struct kvmppc_book3s_shadow_vcpu *svcpu;
-
-				svcpu = svcpu_get(vcpu);
 				*advance = 0;
 				vcpu->arch.shared->dar = vaddr;
-				svcpu->fault_dar = vaddr;
+				vcpu->arch.fault_dar = vaddr;
 
 				dsisr = DSISR_ISSTORE;
 				if (r == -ENOENT)
@@ -281,8 +278,7 @@ int kvmppc_core_emulate_op(struct kvm_run *run, struct kvm_vcpu *vcpu,
 					dsisr |= DSISR_PROTFAULT;
 
 				vcpu->arch.shared->dsisr = dsisr;
-				svcpu->fault_dsisr = dsisr;
-				svcpu_put(svcpu);
+				vcpu->arch.fault_dsisr = dsisr;
 
 				kvmppc_book3s_queue_irqprio(vcpu,
 					BOOK3S_INTERRUPT_DATA_STORAGE);
diff --git a/arch/powerpc/kvm/book3s_interrupts.S b/arch/powerpc/kvm/book3s_interrupts.S
index 17cfae5497a3..d4e30d8fa825 100644
--- a/arch/powerpc/kvm/book3s_interrupts.S
+++ b/arch/powerpc/kvm/book3s_interrupts.S
@@ -26,8 +26,12 @@
 
 #if defined(CONFIG_PPC_BOOK3S_64)
 #define FUNC(name) 		GLUE(.,name)
+#define GET_SHADOW_VCPU(reg)    addi	reg, r13, PACA_SVCPU
+
 #elif defined(CONFIG_PPC_BOOK3S_32)
 #define FUNC(name)		name
+#define GET_SHADOW_VCPU(reg)	lwz     reg, (THREAD + THREAD_KVM_SVCPU)(r2)
+
 #endif /* CONFIG_PPC_BOOK3S_XX */
 
 #define VCPU_LOAD_NVGPRS(vcpu) \
@@ -87,8 +91,14 @@ kvm_start_entry:
 	VCPU_LOAD_NVGPRS(r4)
 
 kvm_start_lightweight:
+	/* Copy registers into shadow vcpu so we can access them in real mode */
+	GET_SHADOW_VCPU(r3)
+	bl	FUNC(kvmppc_copy_to_svcpu)
+	nop
+	REST_GPR(4, r1)
 
 #ifdef CONFIG_PPC_BOOK3S_64
+	/* Get the dcbz32 flag */
 	PPC_LL	r3, VCPU_HFLAGS(r4)
 	rldicl	r3, r3, 0, 63		/* r3 &= 1 */
 	stb	r3, HSTATE_RESTORE_HID5(r13)
@@ -125,18 +135,31 @@ kvmppc_handler_highmem:
 	 *
 	 */
 
-	/* R7 = vcpu */
-	PPC_LL	r7, GPR4(r1)
+	/* Transfer reg values from shadow vcpu back to vcpu struct */
+	/* On 64-bit, interrupts are still off at this point */
+	PPC_LL	r3, GPR4(r1)		/* vcpu pointer */
+	GET_SHADOW_VCPU(r4)
+	bl	FUNC(kvmppc_copy_from_svcpu)
+	nop
 
 #ifdef CONFIG_PPC_BOOK3S_64
+	/* Re-enable interrupts */
+	ld	r3, HSTATE_HOST_MSR(r13)
+	ori	r3, r3, MSR_EE
+	MTMSR_EERI(r3)
+
 	/*
 	 * Reload kernel SPRG3 value.
 	 * No need to save guest value as usermode can't modify SPRG3.
 	 */
 	ld	r3, PACA_SPRG3(r13)
 	mtspr	SPRN_SPRG3, r3
+
 #endif /* CONFIG_PPC_BOOK3S_64 */
 
+	/* R7 = vcpu */
+	PPC_LL	r7, GPR4(r1)
+
 	PPC_STL	r14, VCPU_GPR(R14)(r7)
 	PPC_STL	r15, VCPU_GPR(R15)(r7)
 	PPC_STL	r16, VCPU_GPR(R16)(r7)
diff --git a/arch/powerpc/kvm/book3s_pr.c b/arch/powerpc/kvm/book3s_pr.c
index 8d45f185241a..228a9baffd9e 100644
--- a/arch/powerpc/kvm/book3s_pr.c
+++ b/arch/powerpc/kvm/book3s_pr.c
@@ -61,8 +61,6 @@ void kvmppc_core_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 #ifdef CONFIG_PPC_BOOK3S_64
 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 	memcpy(svcpu->slb, to_book3s(vcpu)->slb_shadow, sizeof(svcpu->slb));
-	memcpy(&get_paca()->shadow_vcpu, to_book3s(vcpu)->shadow_vcpu,
-	       sizeof(get_paca()->shadow_vcpu));
 	svcpu->slb_max = to_book3s(vcpu)->slb_shadow_max;
 	svcpu_put(svcpu);
 #endif
@@ -77,8 +75,6 @@ void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
 #ifdef CONFIG_PPC_BOOK3S_64
 	struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 	memcpy(to_book3s(vcpu)->slb_shadow, svcpu->slb, sizeof(svcpu->slb));
-	memcpy(to_book3s(vcpu)->shadow_vcpu, &get_paca()->shadow_vcpu,
-	       sizeof(get_paca()->shadow_vcpu));
 	to_book3s(vcpu)->slb_shadow_max = svcpu->slb_max;
 	svcpu_put(svcpu);
 #endif
@@ -87,6 +83,60 @@ void kvmppc_core_vcpu_put(struct kvm_vcpu *vcpu)
 	vcpu->cpu = -1;
 }
 
+/* Copy data needed by real-mode code from vcpu to shadow vcpu */
+void kvmppc_copy_to_svcpu(struct kvmppc_book3s_shadow_vcpu *svcpu,
+			  struct kvm_vcpu *vcpu)
+{
+	svcpu->gpr[0] = vcpu->arch.gpr[0];
+	svcpu->gpr[1] = vcpu->arch.gpr[1];
+	svcpu->gpr[2] = vcpu->arch.gpr[2];
+	svcpu->gpr[3] = vcpu->arch.gpr[3];
+	svcpu->gpr[4] = vcpu->arch.gpr[4];
+	svcpu->gpr[5] = vcpu->arch.gpr[5];
+	svcpu->gpr[6] = vcpu->arch.gpr[6];
+	svcpu->gpr[7] = vcpu->arch.gpr[7];
+	svcpu->gpr[8] = vcpu->arch.gpr[8];
+	svcpu->gpr[9] = vcpu->arch.gpr[9];
+	svcpu->gpr[10] = vcpu->arch.gpr[10];
+	svcpu->gpr[11] = vcpu->arch.gpr[11];
+	svcpu->gpr[12] = vcpu->arch.gpr[12];
+	svcpu->gpr[13] = vcpu->arch.gpr[13];
+	svcpu->cr  = vcpu->arch.cr;
+	svcpu->xer = vcpu->arch.xer;
+	svcpu->ctr = vcpu->arch.ctr;
+	svcpu->lr  = vcpu->arch.lr;
+	svcpu->pc  = vcpu->arch.pc;
+}
+
+/* Copy data touched by real-mode code from shadow vcpu back to vcpu */
+void kvmppc_copy_from_svcpu(struct kvm_vcpu *vcpu,
+			    struct kvmppc_book3s_shadow_vcpu *svcpu)
+{
+	vcpu->arch.gpr[0] = svcpu->gpr[0];
+	vcpu->arch.gpr[1] = svcpu->gpr[1];
+	vcpu->arch.gpr[2] = svcpu->gpr[2];
+	vcpu->arch.gpr[3] = svcpu->gpr[3];
+	vcpu->arch.gpr[4] = svcpu->gpr[4];
+	vcpu->arch.gpr[5] = svcpu->gpr[5];
+	vcpu->arch.gpr[6] = svcpu->gpr[6];
+	vcpu->arch.gpr[7] = svcpu->gpr[7];
+	vcpu->arch.gpr[8] = svcpu->gpr[8];
+	vcpu->arch.gpr[9] = svcpu->gpr[9];
+	vcpu->arch.gpr[10] = svcpu->gpr[10];
+	vcpu->arch.gpr[11] = svcpu->gpr[11];
+	vcpu->arch.gpr[12] = svcpu->gpr[12];
+	vcpu->arch.gpr[13] = svcpu->gpr[13];
+	vcpu->arch.cr  = svcpu->cr;
+	vcpu->arch.xer = svcpu->xer;
+	vcpu->arch.ctr = svcpu->ctr;
+	vcpu->arch.lr  = svcpu->lr;
+	vcpu->arch.pc  = svcpu->pc;
+	vcpu->arch.shadow_srr1 = svcpu->shadow_srr1;
+	vcpu->arch.fault_dar   = svcpu->fault_dar;
+	vcpu->arch.fault_dsisr = svcpu->fault_dsisr;
+	vcpu->arch.last_inst   = svcpu->last_inst;
+}
+
 int kvmppc_core_check_requests(struct kvm_vcpu *vcpu)
 {
 	int r = 1; /* Indicate we want to get back into the guest */
@@ -388,22 +438,18 @@ int kvmppc_handle_pagefault(struct kvm_run *run, struct kvm_vcpu *vcpu,
 
 	if (page_found == -ENOENT) {
 		/* Page not found in guest PTE entries */
-		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
-		vcpu->arch.shared->dsisr = svcpu->fault_dsisr;
+		vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr;
 		vcpu->arch.shared->msr |=
-			(svcpu->shadow_srr1 & 0x00000000f8000000ULL);
-		svcpu_put(svcpu);
+			vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
 		kvmppc_book3s_queue_irqprio(vcpu, vec);
 	} else if (page_found == -EPERM) {
 		/* Storage protection */
-		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
 		vcpu->arch.shared->dar = kvmppc_get_fault_dar(vcpu);
-		vcpu->arch.shared->dsisr = svcpu->fault_dsisr & ~DSISR_NOHPTE;
+		vcpu->arch.shared->dsisr = vcpu->arch.fault_dsisr & ~DSISR_NOHPTE;
 		vcpu->arch.shared->dsisr |= DSISR_PROTFAULT;
 		vcpu->arch.shared->msr |=
-			svcpu->shadow_srr1 & 0x00000000f8000000ULL;
-		svcpu_put(svcpu);
+			vcpu->arch.shadow_srr1 & 0x00000000f8000000ULL;
 		kvmppc_book3s_queue_irqprio(vcpu, vec);
 	} else if (page_found == -EINVAL) {
 		/* Page not found in guest SLB */
@@ -645,21 +691,26 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 	switch (exit_nr) {
 	case BOOK3S_INTERRUPT_INST_STORAGE:
 	{
-		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
-		ulong shadow_srr1 = svcpu->shadow_srr1;
+		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 		vcpu->stat.pf_instruc++;
 
 #ifdef CONFIG_PPC_BOOK3S_32
 		/* We set segments as unused segments when invalidating them. So
 		 * treat the respective fault as segment fault. */
-		if (svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT] == SR_INVALID) {
-			kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
-			r = RESUME_GUEST;
+		{
+			struct kvmppc_book3s_shadow_vcpu *svcpu;
+			u32 sr;
+
+			svcpu = svcpu_get(vcpu);
+			sr = svcpu->sr[kvmppc_get_pc(vcpu) >> SID_SHIFT];
 			svcpu_put(svcpu);
-			break;
+			if (sr == SR_INVALID) {
+				kvmppc_mmu_map_segment(vcpu, kvmppc_get_pc(vcpu));
+				r = RESUME_GUEST;
+				break;
+			}
 		}
 #endif
-		svcpu_put(svcpu);
 
 		/* only care about PTEG not found errors, but leave NX alone */
 		if (shadow_srr1 & 0x40000000) {
@@ -684,21 +735,26 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 	case BOOK3S_INTERRUPT_DATA_STORAGE:
 	{
 		ulong dar = kvmppc_get_fault_dar(vcpu);
-		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
-		u32 fault_dsisr = svcpu->fault_dsisr;
+		u32 fault_dsisr = vcpu->arch.fault_dsisr;
 		vcpu->stat.pf_storage++;
 
 #ifdef CONFIG_PPC_BOOK3S_32
 		/* We set segments as unused segments when invalidating them. So
 		 * treat the respective fault as segment fault. */
-		if ((svcpu->sr[dar >> SID_SHIFT]) == SR_INVALID) {
-			kvmppc_mmu_map_segment(vcpu, dar);
-			r = RESUME_GUEST;
+		{
+			struct kvmppc_book3s_shadow_vcpu *svcpu;
+			u32 sr;
+
+			svcpu = svcpu_get(vcpu);
+			sr = svcpu->sr[dar >> SID_SHIFT];
 			svcpu_put(svcpu);
-			break;
+			if (sr == SR_INVALID) {
+				kvmppc_mmu_map_segment(vcpu, dar);
+				r = RESUME_GUEST;
+				break;
+			}
 		}
 #endif
-		svcpu_put(svcpu);
 
 		/* The only case we need to handle is missing shadow PTEs */
 		if (fault_dsisr & DSISR_NOHPTE) {
@@ -745,13 +801,10 @@ int kvmppc_handle_exit(struct kvm_run *run, struct kvm_vcpu *vcpu,
 	case BOOK3S_INTERRUPT_H_EMUL_ASSIST:
 	{
 		enum emulation_result er;
-		struct kvmppc_book3s_shadow_vcpu *svcpu;
 		ulong flags;
 
 program_interrupt:
-		svcpu = svcpu_get(vcpu);
-		flags = svcpu->shadow_srr1 & 0x1f0000ull;
-		svcpu_put(svcpu);
+		flags = vcpu->arch.shadow_srr1 & 0x1f0000ull;
 
 		if (vcpu->arch.shared->msr & MSR_PR) {
 #ifdef EXIT_DEBUG
@@ -883,9 +936,7 @@ program_interrupt:
 		break;
 	default:
 	{
-		struct kvmppc_book3s_shadow_vcpu *svcpu = svcpu_get(vcpu);
-		ulong shadow_srr1 = svcpu->shadow_srr1;
-		svcpu_put(svcpu);
+		ulong shadow_srr1 = vcpu->arch.shadow_srr1;
 		/* Ugh - bork here! What did we get? */
 		printk(KERN_EMERG "exit_nr=0x%x | pc=0x%lx | msr=0x%lx\n",
 			exit_nr, kvmppc_get_pc(vcpu), shadow_srr1);
@@ -1060,11 +1111,12 @@ struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 	if (!vcpu_book3s)
 		goto out;
 
+#ifdef CONFIG_KVM_BOOK3S_32
 	vcpu_book3s->shadow_vcpu =
 		kzalloc(sizeof(*vcpu_book3s->shadow_vcpu), GFP_KERNEL);
 	if (!vcpu_book3s->shadow_vcpu)
 		goto free_vcpu;
-
+#endif
 	vcpu = &vcpu_book3s->vcpu;
 	err = kvm_vcpu_init(vcpu, kvm, id);
 	if (err)
@@ -1098,8 +1150,10 @@ struct kvm_vcpu *kvmppc_core_vcpu_create(struct kvm *kvm, unsigned int id)
 uninit_vcpu:
 	kvm_vcpu_uninit(vcpu);
 free_shadow_vcpu:
+#ifdef CONFIG_KVM_BOOK3S_32
 	kfree(vcpu_book3s->shadow_vcpu);
 free_vcpu:
+#endif
 	vfree(vcpu_book3s);
 out:
 	return ERR_PTR(err);
diff --git a/arch/powerpc/kvm/book3s_rmhandlers.S b/arch/powerpc/kvm/book3s_rmhandlers.S
index 8f7633e3afb8..cd59a3a38482 100644
--- a/arch/powerpc/kvm/book3s_rmhandlers.S
+++ b/arch/powerpc/kvm/book3s_rmhandlers.S
@@ -179,11 +179,15 @@ _GLOBAL(kvmppc_entry_trampoline)
 
 	li	r6, MSR_IR | MSR_DR
 	andc	r6, r5, r6	/* Clear DR and IR in MSR value */
+#ifdef CONFIG_PPC_BOOK3S_32
 	/*
 	 * Set EE in HOST_MSR so that it's enabled when we get into our
-	 * C exit handler function
+	 * C exit handler function.  On 64-bit we delay enabling
+	 * interrupts until we have finished transferring stuff
+	 * to or from the PACA.
 	 */
 	ori	r5, r5, MSR_EE
+#endif
 	mtsrr0	r7
 	mtsrr1	r6
 	RFI
diff --git a/arch/powerpc/kvm/trace.h b/arch/powerpc/kvm/trace.h
index e326489a5420..a088e9a8c103 100644
--- a/arch/powerpc/kvm/trace.h
+++ b/arch/powerpc/kvm/trace.h
@@ -101,17 +101,12 @@ TRACE_EVENT(kvm_exit,
 	),
 
 	TP_fast_assign(
-#ifdef CONFIG_KVM_BOOK3S_PR
-		struct kvmppc_book3s_shadow_vcpu *svcpu;
-#endif
 		__entry->exit_nr	= exit_nr;
 		__entry->pc		= kvmppc_get_pc(vcpu);
 		__entry->dar		= kvmppc_get_fault_dar(vcpu);
 		__entry->msr		= vcpu->arch.shared->msr;
 #ifdef CONFIG_KVM_BOOK3S_PR
-		svcpu = svcpu_get(vcpu);
-		__entry->srr1		= svcpu->shadow_srr1;
-		svcpu_put(svcpu);
+		__entry->srr1		= vcpu->arch.shadow_srr1;
 #endif
 		__entry->last_inst	= vcpu->arch.last_inst;
 	),