Commit c1c6a273 authored by Min Fanlei's avatar Min Fanlei Committed by guzitao
Browse files

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: default avatarMin Fanlei <minfanlei@wxiat.com>
Reviewed-by: default avatarHe Sheng <hesheng@wxiat.com>
Signed-off-by: default avatarGu Zitao <guzitao@wxiat.com>
parent 36b92f77

arch/sw_64/include/asm/kvm_asm.h

+2 −0
Original line number Diff line number Diff line
@@ -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

arch/sw_64/include/asm/vcpu.h

+10 −0
Original line number Diff line number Diff line
@@ -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__ */

arch/sw_64/kvm/Kconfig

+1 −0
Original line number Diff line number Diff line
@@ -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.

arch/sw_64/kvm/handle_exit.c

+9 −0
Original line number Diff line number Diff line
@@ -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);

arch/sw_64/kvm/kvm-sw64.c

+83 −2
Original line number Diff line number Diff line
@@ -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;