Re: [PATCH 1/1] KVM: Add KVM_GET_REG_LIST ioctl for LoongArch

[liushuyu <liushuyu@aosc.io>]

Hi Bibo,
>
>
> On 2026/1/25 下午1:43, Zixing Liu wrote:
>> This ioctl can be used by userspace applications to determine which
>> (special) registers are get/set-able.
>>
>> This can be very useful for cross-platform VMMs so that they do not have
>> to hardcode register indices for each supported architectures.
>>
>> Signed-off-by: Zixing Liu <liushuyu@aosc.io>
>> ---
>>
>> For example, this ioctl could be used by rust-vmm/rust-kvm or maybe
>> VirtualBox-kvm in the future.
>>
>>   Documentation/virt/kvm/api.rst |  2 +-
>>   arch/loongarch/kvm/vcpu.c      | 69 ++++++++++++++++++++++++++++++++++
>>   2 files changed, 70 insertions(+), 1 deletion(-)
>>
>> diff --git a/Documentation/virt/kvm/api.rst
>> b/Documentation/virt/kvm/api.rst
>> index 01a3abef8abb..f46dd8be282f 100644
>> --- a/Documentation/virt/kvm/api.rst
>> +++ b/Documentation/virt/kvm/api.rst
>> @@ -3603,7 +3603,7 @@ VCPU matching underlying host.
>>   ---------------------
>>     :Capability: basic
>> -:Architectures: arm64, mips, riscv, x86 (if KVM_CAP_ONE_REG)
>> +:Architectures: arm64, loongarch, mips, riscv, x86 (if KVM_CAP_ONE_REG)
>>   :Type: vcpu ioctl
>>   :Parameters: struct kvm_reg_list (in/out)
>>   :Returns: 0 on success; -1 on error
>> diff --git a/arch/loongarch/kvm/vcpu.c b/arch/loongarch/kvm/vcpu.c
>> index 656b954c1134..b884eb9c76aa 100644
>> --- a/arch/loongarch/kvm/vcpu.c
>> +++ b/arch/loongarch/kvm/vcpu.c
>> @@ -1186,6 +1186,57 @@ static int kvm_loongarch_vcpu_set_attr(struct
>> kvm_vcpu *vcpu,
>>       return ret;
>>   }
>>   +static unsigned long kvm_loongarch_num_lbt_regs(void)
>> +{
>> +    /* +1 for the LBT_FTOP flag (inside arch.fpu) */
>> +    return sizeof(struct loongarch_lbt) / sizeof(unsigned long) + 1;
>> +}
>> +
>> +static unsigned long kvm_loongarch_num_regs(struct kvm_vcpu *vcpu)
>> +{
>> +    /* +1 for the KVM_REG_LOONGARCH_COUNTER register */
>> +    unsigned long res = CSR_MAX_NUMS + KVM_MAX_CPUCFG_REGS + 1;
>> +
>> +    if (kvm_guest_has_lbt(&vcpu->arch))
>> +        res += kvm_loongarch_num_lbt_regs();
>> +
>> +    return res;
>> +}
>> +
>> +static int kvm_loongarch_copy_reg_indices(struct kvm_vcpu *vcpu,
>> +                      u64 __user *uindices)
>> +{
>> +    u64 reg;
>> +    unsigned int i;
>> +
>> +    for (i = 0; i < CSR_MAX_NUMS; i++) {
>> +        reg = KVM_IOC_CSRID(i);
>> +        if (put_user(reg, uindices++))
>> +            return -EFAULT;
>> +    }
> CSR_MAX_NUMS is max number of accessible CSR registers, instead only
> part of them is used by vCPU model. By my understanding, there will be
> no much meaning if CSR_MAX_NUMS is returned. And I think it will be
> better if real CSR register id and number is returned. 
>
Did you mean we should only return the CSR registers initialized in this
function
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/arch/loongarch/kvm/main.c?id=63804fed149a6750ffd28610c5c1c98cce6bd377#n48
?

That looks like a very large list and the values of the register IDs are
not fully continuous. If that is the case, how do we maintain the
get/set-able CSR register list?

> Regards
> Bibo Mao
>> +
>> +    for (i = 0; i < KVM_MAX_CPUCFG_REGS; i++) {
>> +        reg = KVM_IOC_CPUCFG(i);
>> +        if (put_user(reg, uindices++))
>> +            return -EFAULT;
>> +    }
>> +
>> +    reg = KVM_REG_LOONGARCH_COUNTER;
>> +    if (put_user(reg, uindices++))
>> +        return -EFAULT;
>> +
>> +    if (!kvm_guest_has_lbt(&vcpu->arch))
>> +        return 0;
>> +
>> +    for (i = 1; i <= kvm_loongarch_num_lbt_regs(); i++) {
>> +        reg = (KVM_REG_LOONGARCH_LBT | KVM_REG_SIZE_U64 | i);
>> +        if (put_user(reg, uindices++))
>> +            return -EFAULT;
>> +    }
>> +
>> +    return 0;
>> +}
>> +
>>   long kvm_arch_vcpu_ioctl(struct file *filp,
>>                unsigned int ioctl, unsigned long arg)
>>   {
>> @@ -1251,6 +1302,24 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
>>           r = kvm_loongarch_vcpu_set_attr(vcpu, &attr);
>>           break;
>>       }
>> +    case KVM_GET_REG_LIST: {
>> +        struct kvm_reg_list __user *user_list = argp;
>> +        struct kvm_reg_list reg_list;
>> +        unsigned n;
>> +
>> +        r = -EFAULT;
>> +        if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
>> +            break;
>> +        n = reg_list.n;
>> +        reg_list.n = kvm_loongarch_num_regs(vcpu);
>> +        if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
>> +            break;
>> +        r = -E2BIG;
>> +        if (n < reg_list.n)
>> +            break;
>> +        r = kvm_loongarch_copy_reg_indices(vcpu, user_list->reg);
>> +        break;
>> +    }
>>       default:
>>           r = -ENOIOCTLCMD;
>>           break;
>>
>
Thanks,
Zixing