Re: [PATCH] target/riscv/kvm: Sync vCPU mp_state for migration

["BillXiang" <xiangwencheng@lanxincomputing.com>]

Hi all, sorry for bothering. I just noticed that
Xie Bo <xb@ultrarisc.com> is already working on this.

On 9/16/2025 6:48 PM, BillXiang wrote:
> Fix secondary vCPUs(id > 0) stall after migration.
> 
> Signed-off-by: BillXiang <xiangwencheng@lanxincomputing.com>
> ---
>   target/riscv/cpu.h           |  3 +++
>   target/riscv/kvm/kvm-cpu.c   | 34 ++++++++++++++++++++++++++--------
>   target/riscv/kvm/kvm_riscv.h |  3 ++-
>   target/riscv/machine.c       | 30 ++++++++++++++++++++++++++++++
>   4 files changed, 61 insertions(+), 9 deletions(-)
> 
> diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h
> index 4a862da615..4290229c56 100644
> --- a/target/riscv/cpu.h
> +++ b/target/riscv/cpu.h
> @@ -20,6 +20,7 @@
>   #ifndef RISCV_CPU_H
>   #define RISCV_CPU_H
>   
> +#include <linux/kvm.h>
>   #include "hw/core/cpu.h"
>   #include "hw/registerfields.h"
>   #include "hw/qdev-properties.h"
> @@ -546,6 +547,8 @@ struct ArchCPU {
>       RISCVCPUConfig cfg;
>       RISCVSATPModes satp_modes;
>   
> +    struct kvm_mp_state mp_state;
> +
>       QEMUTimer *pmu_timer;
>       /* A bitmask of Available programmable counters */
>       uint32_t pmu_avail_ctrs;
> diff --git a/target/riscv/kvm/kvm-cpu.c b/target/riscv/kvm/kvm-cpu.c
> index 5c19062c19..f0a97293f9 100644
> --- a/target/riscv/kvm/kvm-cpu.c
> +++ b/target/riscv/kvm/kvm-cpu.c
> @@ -1348,17 +1348,16 @@ int kvm_arch_get_registers(CPUState *cs, Error **errp)
>           return ret;
>       }
>   
> +    RISCVCPU *cpu = RISCV_CPU(cs);
> +    ret = kvm_riscv_sync_mpstate_to_qemu(cpu);
> +
>       return ret;
>   }
>   
> -int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
> +int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu)
>   {
>       if (cap_has_mp_state) {
> -        struct kvm_mp_state mp_state = {
> -            .mp_state = state
> -        };
> -
> -        int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &mp_state);
> +        int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_SET_MP_STATE, &cpu->mp_state);
>           if (ret) {
>               fprintf(stderr, "%s: failed to sync MP_STATE %d/%s\n",
>                       __func__, ret, strerror(-ret));
> @@ -1369,6 +1368,17 @@ int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state)
>       return 0;
>   }
>   
> +int kvm_riscv_sync_mpstate_to_qemu(RISCVCPU *cpu)
> +{
> +    if (cap_has_mp_state) {
> +        int ret = kvm_vcpu_ioctl(CPU(cpu), KVM_GET_MP_STATE, &cpu->mp_state);
> +        if (ret) {
> +            return ret;
> +        }
> +    }
> +    return 0;
> +}
> +
>   int kvm_arch_put_registers(CPUState *cs, int level, Error **errp)
>   {
>       int ret = 0;
> @@ -1396,13 +1406,21 @@ int kvm_arch_put_registers(CPUState *cs, int level, Error **errp)
>       if (KVM_PUT_RESET_STATE == level) {
>           RISCVCPU *cpu = RISCV_CPU(cs);
>           if (cs->cpu_index == 0) {
> -            ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_RUNNABLE);
> +            cpu->mp_state.mp_state = KVM_MP_STATE_RUNNABLE;
> +            ret = kvm_riscv_sync_mpstate_to_kvm(cpu);
>           } else {
> -            ret = kvm_riscv_sync_mpstate_to_kvm(cpu, KVM_MP_STATE_STOPPED);
> +            cpu->mp_state.mp_state = KVM_MP_STATE_STOPPED;
> +            ret = kvm_riscv_sync_mpstate_to_kvm(cpu);
>           }
>           if (ret) {
>               return ret;
>           }
> +    } else if (KVM_PUT_FULL_STATE == level) {
> +        RISCVCPU *cpu = RISCV_CPU(cs);
> +        ret = kvm_riscv_sync_mpstate_to_kvm(cpu);
> +        if (ret) {
> +            return ret;
> +        }
>       }
>   
>       return ret;
> diff --git a/target/riscv/kvm/kvm_riscv.h b/target/riscv/kvm/kvm_riscv.h
> index b2bcd1041f..770f58bf0a 100644
> --- a/target/riscv/kvm/kvm_riscv.h
> +++ b/target/riscv/kvm/kvm_riscv.h
> @@ -28,7 +28,8 @@ void kvm_riscv_aia_create(MachineState *machine, uint64_t group_shift,
>                             uint64_t aplic_base, uint64_t imsic_base,
>                             uint64_t guest_num);
>   void riscv_kvm_aplic_request(void *opaque, int irq, int level);
> -int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu, int state);
> +int kvm_riscv_sync_mpstate_to_kvm(RISCVCPU *cpu);
> +int kvm_riscv_sync_mpstate_to_qemu(RISCVCPU *cpu);
>   void riscv_kvm_cpu_finalize_features(RISCVCPU *cpu, Error **errp);
>   uint64_t kvm_riscv_get_timebase_frequency(RISCVCPU *cpu);
>   
> diff --git a/target/riscv/machine.c b/target/riscv/machine.c
> index 1600ec44f0..178ae6a3a0 100644
> --- a/target/riscv/machine.c
> +++ b/target/riscv/machine.c
> @@ -251,6 +251,28 @@ static const VMStateDescription vmstate_debug = {
>       }
>   };
>   
> +static int get_mp_state(QEMUFile *f, void *opaque, size_t size,
> +                    const VMStateField *field)
> +{
> +    RISCVCPU *cpu = opaque;
> +    cpu->mp_state.mp_state = qemu_get_be32(f);
> +    return 0;
> +}
> +
> +static int put_mp_state(QEMUFile *f, void *opaque, size_t size,
> +                    const VMStateField *field, JSONWriter *vmdesc)
> +{
> +    RISCVCPU *cpu = opaque;
> +    qemu_put_be32(f, cpu->mp_state.mp_state);
> +    return 0;
> +}
> +
> +static const VMStateInfo vmstate_mp_state = {
> +    .name = "mp_state",
> +    .get = get_mp_state,
> +    .put = put_mp_state,
> +};
> +
>   static int riscv_cpu_post_load(void *opaque, int version_id)
>   {
>       RISCVCPU *cpu = opaque;
> @@ -457,6 +479,14 @@ const VMStateDescription vmstate_riscv_cpu = {
>           VMSTATE_UINTTL(env.sscratch, RISCVCPU),
>           VMSTATE_UINTTL(env.mscratch, RISCVCPU),
>           VMSTATE_UINT64(env.stimecmp, RISCVCPU),
> +        {
> +            .name = "mp_state",
> +            .version_id = 0,
> +            .size = sizeof(bool),
> +            .info = &vmstate_mp_state,
> +            .flags = VMS_SINGLE,
> +            .offset = 0,
> +        },
>   
>           VMSTATE_END_OF_LIST()
>       },