sw64: kvm: add guest live migration support

Sunway inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I5XTKO

--------------------------------

This patch adds live migration support for guest os. It requires
hmcode of host and guest to be upgraded to activate this feature.

Signed-off-by: Min Fanlei <minfanlei@wxiat.com>
Reviewed-by: He Sheng <hesheng@wxiat.com>
Signed-off-by: Gu Zitao <guzitao@wxiat.com>

5 files changed, 105 insertions, 2 deletions

diff --git a/arch/sw_64/include/asm/kvm_asm.h b/arch/sw_64/include/asm/kvm_asm.h
index d408e90cee6201..841bfa1dd0aad0 100644
--- a/arch/sw_64/include/asm/kvm_asm.h
+++ b/arch/sw_64/include/asm/kvm_asm.h
@@ -4,6 +4,8 @@
 
 #define SW64_KVM_EXIT_HOST_INTR		0
 #define SW64_KVM_EXIT_IO		1
+#define SW64_KVM_MIGRATION_SET_DIRTY    2
+#define SW64_KVM_MIGRATION_SET_DIRTY_HM 3
 #define SW64_KVM_EXIT_HALT		10
 #define SW64_KVM_EXIT_SHUTDOWN		12
 #define SW64_KVM_EXIT_TIMER		13
diff --git a/arch/sw_64/include/asm/vcpu.h b/arch/sw_64/include/asm/vcpu.h
index dfefb9dc8651ff..c43ebe72e3a1e5 100644
--- a/arch/sw_64/include/asm/vcpu.h
+++ b/arch/sw_64/include/asm/vcpu.h
@@ -41,6 +41,16 @@ struct vcpucb {
 	unsigned long exit_reason;
 	unsigned long ipaddr;
 	unsigned long vcpu_irq_vector;
+	unsigned long pri_base;
+	unsigned long stack_pc_dfault;
+	unsigned long guest_p20;
+	unsigned long guest_dfault_double;
+	unsigned long guest_irqs_pending;
+	unsigned long guest_hm_r30;
+	unsigned long migration_mark;
+	unsigned long guest_longtime;
+	unsigned long guest_longtime_offset;
+	unsigned long reserved[3];
 };
 
 #endif /* __ASSEMBLY__ */
diff --git a/arch/sw_64/kvm/Kconfig b/arch/sw_64/kvm/Kconfig
index 4b6201ff5dc80b..8077ea4527654f 100644
--- a/arch/sw_64/kvm/Kconfig
+++ b/arch/sw_64/kvm/Kconfig
@@ -29,6 +29,7 @@ config KVM
 	select KVM_VFIO
 	select TUN
 	select GENERIC_ALLOCATOR
+	select KVM_GENERIC_DIRTYLOG_READ_PROTECT
 	help
 	  Support for hosting Guest kernels.
 	  We don't support KVM with 3-level page tables yet.
diff --git a/arch/sw_64/kvm/handle_exit.c b/arch/sw_64/kvm/handle_exit.c
index 5016bc0eddc2f8..52f40a4c5803bf 100644
--- a/arch/sw_64/kvm/handle_exit.c
+++ b/arch/sw_64/kvm/handle_exit.c
@@ -13,9 +13,18 @@
 int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *run,
 		int exception_index, struct hcall_args *hargs)
 {
+	gfn_t gfn;
+
 	switch (exception_index) {
 	case SW64_KVM_EXIT_IO:
 		return io_mem_abort(vcpu, run, hargs);
+	case SW64_KVM_MIGRATION_SET_DIRTY_HM:
+	case SW64_KVM_MIGRATION_SET_DIRTY:
+		gfn = hargs->arg2 >> 24;
+		mutex_lock(&vcpu->kvm->slots_lock);
+		kvm_vcpu_mark_page_dirty(vcpu, gfn);
+		mutex_unlock(&vcpu->kvm->slots_lock);
+		return 1;
 	case SW64_KVM_EXIT_HALT:
 		vcpu->arch.halted = 1;
 		kvm_vcpu_block(vcpu);
diff --git a/arch/sw_64/kvm/kvm-sw64.c b/arch/sw_64/kvm/kvm-sw64.c
index 6afff225794750..ffcfdee58a4877 100644
--- a/arch/sw_64/kvm/kvm-sw64.c
+++ b/arch/sw_64/kvm/kvm-sw64.c
@@ -133,6 +133,19 @@ static void sw64_kvm_switch_vpn(struct kvm_vcpu *vcpu)
 	}
 }
 
+static void check_vcpu_requests(struct kvm_vcpu *vcpu)
+{
+	unsigned long vpn;
+	long cpu = smp_processor_id();
+
+	if (kvm_request_pending(vcpu)) {
+		if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
+			vpn = vcpu->arch.vpnc[cpu] & HARDWARE_VPN_MASK;
+			tbivpn(0, 0, vpn);
+		}
+	}
+}
+
 struct kvm_stats_debugfs_item debugfs_entries[] = {
 	{ NULL }
 };
@@ -177,12 +190,47 @@ int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
 	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
 }
 
+/*
+ * kvm_mark_migration write the mark on every vcpucbs of the kvm, which tells
+ * the system to do migration while the mark is on, and flush all vcpu's tlbs
+ * at the beginning of the migration.
+ */
+void kvm_mark_migration(struct kvm *kvm, int mark)
+{
+	struct kvm_vcpu *vcpu;
+	int cpu;
+
+	kvm_for_each_vcpu(cpu, vcpu, kvm)
+		vcpu->arch.vcb.migration_mark = mark << 2;
+
+	kvm_flush_remote_tlbs(kvm);
+}
+
 void kvm_arch_commit_memory_region(struct kvm *kvm,
 		const struct kvm_userspace_memory_region *mem,
 		struct kvm_memory_slot *old,
 		const struct kvm_memory_slot *new,
 		enum kvm_mr_change change)
 {
+	/*
+	 * At this point memslot has been committed and there is an
+	 * allocated dirty_bitmap[], dirty pages will be be tracked while the
+	 * memory slot is write protected.
+	 */
+
+	/* If dirty logging has been stopped, do nothing for now. */
+	if ((change != KVM_MR_DELETE)
+			&& (old->flags & KVM_MEM_LOG_DIRTY_PAGES)
+			&& (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))) {
+		kvm_mark_migration(kvm, 0);
+		return;
+	}
+
+	/* If it's the first time dirty logging, flush all vcpu tlbs. */
+	if ((change == KVM_MR_FLAGS_ONLY)
+			&& (!(old->flags & KVM_MEM_LOG_DIRTY_PAGES))
+			&& (new->flags & KVM_MEM_LOG_DIRTY_PAGES))
+		kvm_mark_migration(kvm, 1);
 }
 
 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
@@ -193,6 +241,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_IRQCHIP:
 	case KVM_CAP_IOEVENTFD:
 	case KVM_CAP_SYNC_MMU:
+	case KVM_CAP_IMMEDIATE_EXIT:
 		r = 1;
 		break;
 	case KVM_CAP_NR_VCPUS:
@@ -206,9 +255,10 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	return r;
 }
 
-int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
+void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
+		struct kvm_memory_slot *slot, gfn_t gfn_offset,
+		unsigned long mask)
 {
-	return 0;
 }
 
 int kvm_sw64_pending_timer(struct kvm_vcpu *vcpu)
@@ -547,6 +597,9 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
 	bool more;
 	sigset_t sigsaved;
 
+	if (run->immediate_exit)
+		return -EINTR;
+
 	/* Set guest vcb */
 	/* vpn will update later when vcpu is running */
 	if (vcpu->arch.vcb.vpcr == 0) {
@@ -613,6 +666,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
 		vcpu->arch.halted = 0;
 
 		sw64_kvm_switch_vpn(vcpu);
+		check_vcpu_requests(vcpu);
 		guest_enter_irqoff();
 
 		/* Enter the guest */
@@ -647,6 +701,7 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
 long kvm_arch_vcpu_ioctl(struct file *filp,
 		unsigned int ioctl, unsigned long arg)
 {
+	unsigned long result;
 	struct kvm_vcpu *vcpu = filp->private_data;
 	struct vcpucb *kvm_vcb;
 
@@ -654,12 +709,32 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
 	case KVM_SW64_VCPU_INIT:
 		return kvm_arch_vcpu_reset(vcpu);
 	case KVM_SW64_GET_VCB:
+		if (vcpu->arch.vcb.migration_mark) {
+			result = sw64_io_read(0, LONG_TIME);
+			vcpu->arch.vcb.guest_longtime = result;
+			vcpu->arch.vcb.guest_irqs_pending = vcpu->arch.irqs_pending[0];
+		}
+
 		if (copy_to_user((void __user *)arg, &(vcpu->arch.vcb), sizeof(struct vcpucb)))
 			return -EINVAL;
 		break;
 	case KVM_SW64_SET_VCB:
 		kvm_vcb = memdup_user((void __user *)arg, sizeof(*kvm_vcb));
 		memcpy(&(vcpu->arch.vcb), kvm_vcb, sizeof(struct vcpucb));
+
+		if (vcpu->arch.vcb.migration_mark) {
+			/* updated vpcr needed by destination vm */
+			vcpu->arch.vcb.vpcr
+				= get_vpcr(vcpu->kvm->arch.host_phys_addr, vcpu->kvm->arch.size, 0);
+
+			result = sw64_io_read(0, LONG_TIME);
+
+			/* synchronize the longtime of source and destination */
+			vcpu->arch.vcb.guest_longtime_offset = vcpu->arch.vcb.guest_longtime - result;
+
+			set_timer(vcpu, 200000000);
+			vcpu->arch.vcb.migration_mark = 0;
+		}
 		break;
 	default:
 		return -EINVAL;
@@ -702,6 +777,12 @@ void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
 {
 }
 
+void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
+					struct kvm_memory_slot *memslot)
+{
+	kvm_flush_remote_tlbs(kvm);
+}
+
 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
 {
 	return 0;