KVM: PPC: Book3S HV: Virtualize doorbell facility on POWER9

On POWER9, we no longer have the restriction that we had on POWER8
where all threads in a core have to be in the same partition, so
the CPU threads are now independent.  However, we still want to be
able to run guests with a virtual SMT topology, if only to allow
migration of guests from POWER8 systems to POWER9.

A guest that has a virtual SMT mode greater than 1 will expect to
be able to use the doorbell facility; it will expect the msgsndp
and msgclrp instructions to work appropriately and to be able to read
sensible values from the TIR (thread identification register) and
DPDES (directed privileged doorbell exception status) special-purpose
registers.  However, since each CPU thread is a separate sub-processor
in POWER9, these instructions and registers can only be used within
a single CPU thread.

In order for these instructions to appear to act correctly according
to the guest's virtual SMT mode, we have to trap and emulate them.
We cause them to trap by clearing the HFSCR_MSGP bit in the HFSCR
register.  The emulation is triggered by the hypervisor facility
unavailable interrupt that occurs when the guest uses them.

To cause a doorbell interrupt to occur within the guest, we set the
DPDES register to 1.  If the guest has interrupts enabled, the CPU
will generate a doorbell interrupt and clear the DPDES register in
hardware.  The DPDES hardware register for the guest is saved in the
vcpu->arch.vcore->dpdes field.  Since this gets written by the guest
exit code, other VCPUs wishing to cause a doorbell interrupt don't
write that field directly, but instead set a vcpu->arch.doorbell_request
flag.  This is consumed and set to 0 by the guest entry code, which
then sets DPDES to 1.

Emulating reads of the DPDES register is somewhat involved, because
it requires reading the doorbell pending interrupt status of all of the
VCPU threads in the virtual core, and if any of those VCPUs are
running, their doorbell status is only up-to-date in the hardware
DPDES registers of the CPUs where they are running.  In order to get
a reasonable approximation of the current doorbell status, we send
those CPUs an IPI, causing an exit from the guest which will update
the vcpu->arch.vcore->dpdes field.  We then use that value in
constructing the emulated DPDES register value.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

arch/powerpc/include/asm/kvm_host.h

@@ -268,6 +268,7 @@ struct kvm_resize_hpt;
 struct kvm_arch {
 	unsigned int lpid;
 	unsigned int smt_mode;		/* # vcpus per virtual core */
+	unsigned int emul_smt_mode;	/* emualted SMT mode, on P9 */
 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 	unsigned int tlb_sets;
 	struct kvm_hpt_info hpt;
@@ -712,6 +713,7 @@ struct kvm_vcpu_arch {
 	unsigned long pending_exceptions;
 	u8 ceded;
 	u8 prodded;
+	u8 doorbell_request;
 	u32 last_inst;
 
 	struct swait_queue_head *wqp;

arch/powerpc/include/asm/ppc-opcode.h

@@ -103,6 +103,8 @@
 #define OP_31_XOP_STBUX     247
 #define OP_31_XOP_LHZX      279
 #define OP_31_XOP_LHZUX     311
+#define OP_31_XOP_MSGSNDP   142
+#define OP_31_XOP_MSGCLRP   174
 #define OP_31_XOP_MFSPR     339
 #define OP_31_XOP_LWAX      341
 #define OP_31_XOP_LHAX      343

arch/powerpc/kernel/asm-offsets.c

@@ -513,6 +513,7 @@ int main(void)
 	OFFSET(VCPU_PENDING_EXC, kvm_vcpu, arch.pending_exceptions);
 	OFFSET(VCPU_CEDED, kvm_vcpu, arch.ceded);
 	OFFSET(VCPU_PRODDED, kvm_vcpu, arch.prodded);
+	OFFSET(VCPU_DBELL_REQ, kvm_vcpu, arch.doorbell_request);
 	OFFSET(VCPU_MMCR, kvm_vcpu, arch.mmcr);
 	OFFSET(VCPU_PMC, kvm_vcpu, arch.pmc);
 	OFFSET(VCPU_SPMC, kvm_vcpu, arch.spmc);

arch/powerpc/kvm/book3s_hv.c

@@ -46,6 +46,8 @@
 #include <linux/of.h>
 
 #include <asm/reg.h>
+#include <asm/ppc-opcode.h>
+#include <asm/disassemble.h>
 #include <asm/cputable.h>
 #include <asm/cacheflush.h>
 #include <asm/tlbflush.h>
@@ -681,6 +683,15 @@ static bool kvmppc_doorbell_pending(struct kvm_vcpu *vcpu)
 	int thr;
 	struct kvmppc_vcore *vc;
 
+	if (vcpu->arch.doorbell_request)
+		return true;
+	/*
+	 * Ensure that the read of vcore->dpdes comes after the read
+	 * of vcpu->doorbell_request.  This barrier matches the
+	 * lwsync in book3s_hv_rmhandlers.S just before the
+	 * fast_guest_return label.
+	 */
+	smp_rmb();
 	vc = vcpu->arch.vcore;
 	thr = vcpu->vcpu_id - vc->first_vcpuid;
 	return !!(vc->dpdes & (1 << thr));
@@ -937,6 +948,101 @@ static int kvmppc_emulate_debug_inst(struct kvm_run *run,
 	}
 }
 
+static void do_nothing(void *x)
+{
+}
+
+static unsigned long kvmppc_read_dpdes(struct kvm_vcpu *vcpu)
+{
+	int thr, cpu, pcpu, nthreads;
+	struct kvm_vcpu *v;
+	unsigned long dpdes;
+
+	nthreads = vcpu->kvm->arch.emul_smt_mode;
+	dpdes = 0;
+	cpu = vcpu->vcpu_id & ~(nthreads - 1);
+	for (thr = 0; thr < nthreads; ++thr, ++cpu) {
+		v = kvmppc_find_vcpu(vcpu->kvm, cpu);
+		if (!v)
+			continue;
+		/*
+		 * If the vcpu is currently running on a physical cpu thread,
+		 * interrupt it in order to pull it out of the guest briefly,
+		 * which will update its vcore->dpdes value.
+		 */
+		pcpu = READ_ONCE(v->cpu);
+		if (pcpu >= 0)
+			smp_call_function_single(pcpu, do_nothing, NULL, 1);
+		if (kvmppc_doorbell_pending(v))
+			dpdes |= 1 << thr;
+	}
+	return dpdes;
+}
+
+/*
+ * On POWER9, emulate doorbell-related instructions in order to
+ * give the guest the illusion of running on a multi-threaded core.
+ * The instructions emulated are msgsndp, msgclrp, mfspr TIR,
+ * and mfspr DPDES.
+ */
+static int kvmppc_emulate_doorbell_instr(struct kvm_vcpu *vcpu)
+{
+	u32 inst, rb, thr;
+	unsigned long arg;
+	struct kvm *kvm = vcpu->kvm;
+	struct kvm_vcpu *tvcpu;
+
+	if (!cpu_has_feature(CPU_FTR_ARCH_300))
+		return EMULATE_FAIL;
+	if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &inst) != EMULATE_DONE)
+		return RESUME_GUEST;
+	if (get_op(inst) != 31)
+		return EMULATE_FAIL;
+	rb = get_rb(inst);
+	thr = vcpu->vcpu_id & (kvm->arch.emul_smt_mode - 1);
+	switch (get_xop(inst)) {
+	case OP_31_XOP_MSGSNDP:
+		arg = kvmppc_get_gpr(vcpu, rb);
+		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
+			break;
+		arg &= 0x3f;
+		if (arg >= kvm->arch.emul_smt_mode)
+			break;
+		tvcpu = kvmppc_find_vcpu(kvm, vcpu->vcpu_id - thr + arg);
+		if (!tvcpu)
+			break;
+		if (!tvcpu->arch.doorbell_request) {
+			tvcpu->arch.doorbell_request = 1;
+			kvmppc_fast_vcpu_kick_hv(tvcpu);
+		}
+		break;
+	case OP_31_XOP_MSGCLRP:
+		arg = kvmppc_get_gpr(vcpu, rb);
+		if (((arg >> 27) & 0xf) != PPC_DBELL_SERVER)
+			break;
+		vcpu->arch.vcore->dpdes = 0;
+		vcpu->arch.doorbell_request = 0;
+		break;
+	case OP_31_XOP_MFSPR:
+		switch (get_sprn(inst)) {
+		case SPRN_TIR:
+			arg = thr;
+			break;
+		case SPRN_DPDES:
+			arg = kvmppc_read_dpdes(vcpu);
+			break;
+		default:
+			return EMULATE_FAIL;
+		}
+		kvmppc_set_gpr(vcpu, get_rt(inst), arg);
+		break;
+	default:
+		return EMULATE_FAIL;
+	}
+	kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
+	return RESUME_GUEST;
+}
+
 static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 				 struct task_struct *tsk)
 {
@@ -1059,12 +1165,19 @@ static int kvmppc_handle_exit_hv(struct kvm_run *run, struct kvm_vcpu *vcpu,
 		break;
 	/*
 	 * This occurs if the guest (kernel or userspace), does something that
-	 * is prohibited by HFSCR.  We just generate a program interrupt to
-	 * the guest.
+	 * is prohibited by HFSCR.
+	 * On POWER9, this could be a doorbell instruction that we need
+	 * to emulate.
+	 * Otherwise, we just generate a program interrupt to the guest.
 	 */
 	case BOOK3S_INTERRUPT_H_FAC_UNAVAIL:
-		kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
-		r = RESUME_GUEST;
+		r = EMULATE_FAIL;
+		if ((vcpu->arch.hfscr >> 56) == FSCR_MSGP_LG)
+			r = kvmppc_emulate_doorbell_instr(vcpu);
+		if (r == EMULATE_FAIL) {
+			kvmppc_core_queue_program(vcpu, SRR1_PROGILL);
+			r = RESUME_GUEST;
+		}
 		break;
 	case BOOK3S_INTERRUPT_HV_RM_HARD:
 		r = RESUME_PASSTHROUGH;
@@ -1826,10 +1939,14 @@ static struct kvm_vcpu *kvmppc_core_vcpu_create_hv(struct kvm *kvm,
 	 * This value is only used on POWER9.
 	 * On POWER9 DD1, TM doesn't work, so we make sure to
 	 * prevent the guest from using it.
+	 * On POWER9, we want to virtualize the doorbell facility, so we
+	 * turn off the HFSCR bit, which causes those instructions to trap.
 	 */
 	vcpu->arch.hfscr = mfspr(SPRN_HFSCR);
 	if (!cpu_has_feature(CPU_FTR_TM))
 		vcpu->arch.hfscr &= ~HFSCR_TM;
+	if (cpu_has_feature(CPU_FTR_ARCH_300))
+		vcpu->arch.hfscr &= ~HFSCR_MSGP;
 
 	kvmppc_mmu_book3s_hv_init(vcpu);
 
@@ -1880,6 +1997,7 @@ static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
 			      unsigned long flags)
 {
 	int err;
+	int esmt = 0;
 
 	if (flags)
 		return -EINVAL;
@@ -1897,12 +2015,14 @@ static int kvmhv_set_smt_mode(struct kvm *kvm, unsigned long smt_mode,
 		 * On POWER9, the threading mode is "loose",
 		 * so each vcpu gets its own vcore.
 		 */
+		esmt = smt_mode;
 		smt_mode = 1;
 	}
 	mutex_lock(&kvm->lock);
 	err = -EBUSY;
 	if (!kvm->arch.online_vcores) {
 		kvm->arch.smt_mode = smt_mode;
+		kvm->arch.emul_smt_mode = esmt;
 		err = 0;
 	}
 	mutex_unlock(&kvm->lock);
@@ -2025,10 +2145,6 @@ static void kvmppc_release_hwthread(int cpu)
 	tpaca->kvm_hstate.kvm_split_mode = NULL;
 }
 
-static void do_nothing(void *x)
-{
-}
-
 static void radix_flush_cpu(struct kvm *kvm, int cpu, struct kvm_vcpu *vcpu)
 {
 	int i;
@@ -3600,6 +3716,7 @@ static int kvmppc_core_init_vm_hv(struct kvm *kvm)
 		kvm->arch.smt_mode = threads_per_subcore;
 	else
 		kvm->arch.smt_mode = 1;
+	kvm->arch.emul_smt_mode = 1;
 
 	/*
 	 * Create a debugfs directory for the VM

arch/powerpc/kvm/book3s_hv_rmhandlers.S

@@ -1069,6 +1069,23 @@ END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
 	mr	r9, r4
 	bl	kvmppc_msr_interrupt
 5:
+BEGIN_FTR_SECTION
+	b	fast_guest_return
+END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
+	/* On POWER9, check for pending doorbell requests */
+	lbz	r0, VCPU_DBELL_REQ(r4)
+	cmpwi	r0, 0
+	beq	fast_guest_return
+	ld	r5, HSTATE_KVM_VCORE(r13)
+	/* Set DPDES register so the CPU will take a doorbell interrupt */
+	li	r0, 1
+	mtspr	SPRN_DPDES, r0
+	std	r0, VCORE_DPDES(r5)
+	/* Make sure other cpus see vcore->dpdes set before dbell req clear */
+	lwsync
+	/* Clear the pending doorbell request */
+	li	r0, 0
+	stb	r0, VCPU_DBELL_REQ(r4)
 
 /*
  * Required state:

arch/powerpc/kvm/powerpc.c

@@ -554,9 +554,12 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 	case KVM_CAP_PPC_SMT:
 		r = 0;
-		if (kvm)
-			r = kvm->arch.smt_mode;
-		else if (hv_enabled) {
+		if (kvm) {
+			if (kvm->arch.emul_smt_mode > 1)
+				r = kvm->arch.emul_smt_mode;
+			else
+				r = kvm->arch.smt_mode;
+		} else if (hv_enabled) {
 			if (cpu_has_feature(CPU_FTR_ARCH_300))
 				r = 1;
 			else