HVM-217 hat_pfnum call results in mutex_enter: adaptive at high PIL panic

7 files changed, 117 insertions, 76 deletions

diff --git a/kvm.c b/kvm.c
index 0e054f8..ac96d15 100644
--- a/kvm.c
+++ b/kvm.c
@@ -2251,11 +2251,7 @@ kvm_mmu_zap_page(struct kvm *kvm, struct kvm_mmu_page *sp)
 		sp->role.invalid = 1;
 		if (!list_link_active(&sp->link))
 			list_insert_head(&kvm->arch.active_mmu_pages, sp);
-#ifdef XXX
 		kvm_reload_remote_mmus(kvm);
-#else
-		XXX_KVM_PROBE;
-#endif
 	}
 	kvm_mmu_reset_last_pte_updated(kvm);
 
@@ -4644,20 +4640,18 @@ out:
 #define	__ex(x) __kvm_handle_fault_on_reboot(x)
 
 void
-vmcs_clear(struct vmcs *vmcs)
+vmcs_clear(uint64_t vmcs_pa)
 {
 	unsigned char error;
-	uint64_t phys_addr = (hat_getpfnum(kas.a_hat, (caddr_t)vmcs) <<
-	    PAGESHIFT) | ((uint64_t)vmcs & PAGEOFFSET);
 
 	/*CSTYLED*/
 	__asm__ volatile (__ex(ASM_VMX_VMCLEAR_RAX) "\n\tsetna %0\n"
-	    : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
+	    : "=g"(error) : "a"(&vmcs_pa), "m"(vmcs_pa)
 	    : "cc", "memory");
 
 	if (error)
-		cmn_err(CE_PANIC, "kvm: vmclear fail: %p/%lx\n",
-			vmcs, phys_addr);
+		cmn_err(CE_PANIC, "kvm: vmclear fail: %lx\n",
+			vmcs_pa);
 }
 
 void
@@ -4669,7 +4663,7 @@ __vcpu_clear(void *arg)
 	vmx->vmcs->revision_id = vmcs_config.revision_id;
 
 	if (vmx->vcpu.cpu == cpu)
-		vmcs_clear(vmx->vmcs);
+		vmcs_clear(vmx->vmcs_pa);
 
 	if (current_vmcs[cpu] == vmx->vmcs)
 		current_vmcs[cpu] = NULL;
@@ -5019,6 +5013,11 @@ vmx_vcpu_setup(struct vcpu_vmx *vmx)
 	return (0);
 }
 
+static void kvm_migrate_timers(struct kvm_vcpu *vcpu)
+{
+	set_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests);
+}
+
 /*
  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
  * vcpu mutex is already taken.
@@ -5033,11 +5032,7 @@ vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 
 	if (vcpu->cpu != cpu) {
 		vcpu_clear(vmx);
-#ifdef XXX
 		kvm_migrate_timers(vcpu);
-#else
-		XXX_KVM_PROBE;
-#endif
 		set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
 #ifdef XXX
 		kpreempt_disable();
@@ -12007,11 +12002,7 @@ vcpu_enter_guest(struct kvm_vcpu *vcpu)
 	if (vcpu->requests) {
 		if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER,
 		    &vcpu->requests)) {
-#ifdef XXX
 			__kvm_migrate_timers(vcpu);
-#else
-			XXX_KVM_PROBE;
-#endif
 		}
 		if (test_and_clear_bit(KVM_REQ_KVMCLOCK_UPDATE,
 		    &vcpu->requests)) {
diff --git a/kvm.h b/kvm.h
index 3f2ac61..dfe7280 100644
--- a/kvm.h
+++ b/kvm.h
@@ -1818,6 +1818,8 @@ typedef struct vcpu_vmx {
 	uint64_t 		      msr_guest_kernel_gs_base;
 #endif
 	struct vmcs          *vmcs;
+	uint64_t	vmcs_pa;  /* physical address of vmcs for this vmx */
+
 	struct {
 		int           loaded;
 		unsigned short           fs_sel, gs_sel, ldt_sel;
diff --git a/kvm_i8254.c b/kvm_i8254.c
index 78fef5a..bc8fd1b 100644
--- a/kvm_i8254.c
+++ b/kvm_i8254.c
@@ -755,32 +755,6 @@ fail:
 	return (NULL);
 }
 
-void
-kvm_free_pit(struct kvm *kvmp)
-{
-	struct kvm_timer *kptp;
-
-	if (kvmp->arch.vpit == NULL)
-		return;
-
-	mutex_enter(&kvmp->arch.vpit->pit_state.lock);
-	kvm_unregister_irq_mask_notifier(kvmp, 0,
-	    &kvmp->arch.vpit->mask_notifier);
-	kvm_unregister_irq_ack_notifier(kvmp,
-	    &kvmp->arch.vpit->pit_state.irq_ack_notifier);
-	mutex_exit(&kvmp->arch.vpit->pit_state.lock);
-
-	mutex_enter(&cpu_lock);
-	kptp = &kvmp->arch.vpit->pit_state.pit_timer;
-	if (kptp->active)
-		cyclic_remove(kptp->kvm_cyclic_id);
-	mutex_exit(&cpu_lock);
-	mutex_destroy(&kvmp->arch.vpit->pit_state.lock);
-	kvm_free_irq_source_id(kvmp, kvmp->arch.vpit->irq_source_id);
-	kmem_free(kvmp->arch.vpit, sizeof (struct kvm_pit));
-	kvmp->arch.vpit = NULL;
-}
-
 static void
 __inject_pit_timer_intr(struct kvm *kvm)
 {
@@ -839,3 +813,47 @@ kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu)
 			__inject_pit_timer_intr(kvm);
 	}
 }
+
+void
+kvm_free_pit(struct kvm *kvmp)
+{
+	struct kvm_timer *kptp;
+
+	if (kvmp->arch.vpit == NULL)
+		return;
+
+	mutex_enter(&kvmp->arch.vpit->pit_state.lock);
+	kvm_unregister_irq_mask_notifier(kvmp, 0,
+	    &kvmp->arch.vpit->mask_notifier);
+	kvm_unregister_irq_ack_notifier(kvmp,
+	    &kvmp->arch.vpit->pit_state.irq_ack_notifier);
+	mutex_exit(&kvmp->arch.vpit->pit_state.lock);
+
+	mutex_enter(&cpu_lock);
+	kptp = &kvmp->arch.vpit->pit_state.pit_timer;
+	if (kptp->active)
+		cyclic_remove(kptp->kvm_cyclic_id);
+	mutex_exit(&cpu_lock);
+	mutex_destroy(&kvmp->arch.vpit->pit_state.lock);
+	kvm_free_irq_source_id(kvmp, kvmp->arch.vpit->irq_source_id);
+	kmem_free(kvmp->arch.vpit, sizeof (struct kvm_pit));
+	kvmp->arch.vpit = NULL;
+}
+
+void
+__kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
+{
+#ifdef XXX
+	struct kvm_pit *pit = vcpu->kvm->arch.vpit;
+	struct hrtimer *timer;
+
+	if (!kvm_vcpu_is_bsp(vcpu) || !pit)
+		return;
+
+	timer = &pit->pit_state.pit_timer.timer;
+	if (hrtimer_cancel_p(timer))
+		kvm_hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
+#else
+	XXX_KVM_PROBE;
+#endif
+}
diff --git a/kvm_irq.c b/kvm_irq.c
index 7d698b6..d943ac9 100644
--- a/kvm_irq.c
+++ b/kvm_irq.c
@@ -91,3 +91,10 @@ kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)
 unlock:
 	mutex_exit(&kvm->irq_lock);
 }
+
+void
+__kvm_migrate_timers(struct kvm_vcpu *vcpu)
+{
+	__kvm_migrate_apic_timer(vcpu);
+	__kvm_migrate_pit_timer(vcpu);
+}
diff --git a/kvm_lapic.c b/kvm_lapic.c
index 3253137..54db1e6 100644
--- a/kvm_lapic.c
+++ b/kvm_lapic.c
@@ -122,3 +122,21 @@ kvm_free_lapic(struct kvm_vcpu *vcpu)
 
 	kmem_free(vcpu->arch.apic, sizeof (struct kvm_lapic));
 }
+
+void
+__kvm_migrate_apic_timer(struct kvm_vcpu *vcpu)
+{
+#ifdef XXX
+	struct kvm_lapic *apic = vcpu->arch.apic;
+	struct hrtimer *timer;
+
+	if (!apic)
+		return;
+
+	timer = &apic->lapic_timer.timer;
+	if (hrtimer_cancel_p(timer))
+		kvm_hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
+#else
+	XXX_KVM_PROBE;
+#endif
+}
diff --git a/kvm_subr.c b/kvm_subr.c
index 44f045c..b384a29 100644
--- a/kvm_subr.c
+++ b/kvm_subr.c
@@ -184,7 +184,7 @@ unsigned long kvm_read_tr_base(void)
 	return segment_base(tr);
 }
 
-static int
+int
 kvm_xcall_func(kvm_xcall_t func, void *arg)
 {
 	(*func)(arg);
diff --git a/kvm_x86.c b/kvm_x86.c
index 0b8ccc6..d678552 100644
--- a/kvm_x86.c
+++ b/kvm_x86.c
@@ -22,6 +22,7 @@
 #include <sys/mman.h>
 #include <sys/mach_mmu.h>
 #include <sys/int_limits.h>
+#include <sys/x_call.h>
 
 #include "msr-index.h"
 #include "msr.h"
@@ -1714,7 +1715,7 @@ extern int enable_ept;
 extern int enable_unrestricted_guest;
 extern int emulate_invalid_guest_state;
 
-extern void vmcs_clear(struct vmcs *vmcs);
+extern void vmcs_clear(uint64_t vmcs_pa);
 extern void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
 extern void vmx_vcpu_put(struct kvm_vcpu *vcpu);
 
@@ -1835,6 +1836,9 @@ vmx_create_vcpu(struct kvm *kvm, unsigned int id)
 		return (NULL);
 	}
 
+	vmx->vmcs_pa = (hat_getpfnum(kas.a_hat, (caddr_t)vmx->vmcs) <<
+	    PAGESHIFT) | ((int64_t)(vmx->vmcs) & 0xfff);
+
 	kpreempt_disable();
 
 	cpu = curthread->t_cpu->cpu_seqid;
@@ -1842,7 +1846,7 @@ vmx_create_vcpu(struct kvm *kvm, unsigned int id)
 	cmn_err(CE_NOTE, "vmcs revision_id = %x\n", vmcs_config.revision_id);
 	vmx->vmcs->revision_id = vmcs_config.revision_id;
 
-	vmcs_clear(vmx->vmcs);
+	vmcs_clear(vmx->vmcs_pa);
 
 	vmx_vcpu_load(&vmx->vcpu, cpu);
 	err = vmx_vcpu_setup(vmx);
@@ -3872,51 +3876,52 @@ kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
 	return (0);
 }
 
+extern void kvm_xcall(processorid_t cpu, kvm_xcall_t func, void *arg);
+extern int kvm_xcall_func(kvm_xcall_t func, void *arg);
+
+static void
+ack_flush(void *_completed)
+{
+}
+
+extern int kvm_xcall_func(kvm_xcall_t func, void *arg);
+
 int
 make_all_cpus_request(struct kvm *kvm, unsigned int req)
 {
-	int i, cpu, me;
+	int i;
+	cpuset_t set;
+	processorid_t me, cpu;
 #ifdef XXX_KVM_DECLARATION
 	cpumask_var_t cpus;
 #endif
-	int called = 1;
+	int called = 0;
 	struct kvm_vcpu *vcpu;
 
-#ifdef XXX
-	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
+	CPUSET_ZERO(set);
+
 	mutex_enter(&kvm->requests_lock);
-	me = smp_processor_id();
-	kvm_for_each_vcpu(i, vcpu, kvm) {
-#else
-	XXX_KVM_PROBE;
+	me = curthread->t_cpu->cpu_id;
 	for (i = 0; i < 1; i++) {
 		vcpu = kvm->vcpus[i];
 		if (!vcpu)
-			return (0);
-#endif
-
+			break;
 		if (test_and_set_bit(req, &vcpu->requests))
 			continue;
 		cpu = vcpu->cpu;
-#ifdef XXX
-		if (cpus != NULL && cpu != -1 && cpu != me)
-			cpumask_set_cpu(cpu, cpus);
-#else
-		XXX_KVM_PROBE;
-#endif
+		if (cpu != -1 && cpu != me)
+			CPUSET_ADD(set, cpu);
+	}
+	if (CPUSET_ISNULL(set))
+		kvm_xcall(KVM_CPUALL, ack_flush, NULL);
+	else {
+		kpreempt_disable();
+		xc_sync((xc_arg_t) ack_flush, (xc_arg_t) NULL,
+			0, CPUSET2BV(set), (xc_func_t) kvm_xcall_func);
+		kpreempt_enable();
 	}
-#ifdef XXX
-	if (unlikely(cpus == NULL))
-		smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
-	else if (!cpumask_empty(cpus))
-		smp_call_function_many(cpus, ack_flush, NULL, 1);
-	else
-		called = false;
 	mutex_exit(&kvm->requests_lock);
-	free_cpumask_var(cpus);
-#else
-	XXX_KVM_PROBE;
-#endif
+	called = 1;
 
 	return (called);
 }