Re: [PATCH] target/loongarch/kvm: set pv features per vCPU in kvm_arch_init_vcpu

[Tao Cui <cui.tao@linux.dev>]

在 2026/6/25 11:14, Bibo Mao 写道:
> 
> 
> On 2026/6/25 上午9:53, Tao Cui wrote:
>> From: Tao Cui <cuitao@kylinos.cn>
>>
>> kvm_set_pv_features() programs the KVM_FEATURE cpucfg attribute, which is a
>> per-vCPU setting. It was called from kvm_arch_put_registers() under a
>> function-local static guard, so it ran only once for the whole VM and only
>> the first vCPU actually got its pv features pushed to KVM.
>>
>> Move the call to kvm_arch_init_vcpu(), which runs once per vCPU, right
>> after the pv features are computed, and drop the static guard.
>>
>> Signed-off-by: Tao Cui <cuitao@kylinos.cn>
>> ---
>>   target/loongarch/kvm/kvm.c | 15 ++++++---------
>>   1 file changed, 6 insertions(+), 9 deletions(-)
>>
>> diff --git a/target/loongarch/kvm/kvm.c b/target/loongarch/kvm/kvm.c
>> index d6539c12ac..985812c828 100644
>> --- a/target/loongarch/kvm/kvm.c
>> +++ b/target/loongarch/kvm/kvm.c
>> @@ -816,7 +816,6 @@ int kvm_arch_get_registers(CPUState *cs, Error **errp)
>>   int kvm_arch_put_registers(CPUState *cs, KvmPutState level, Error **errp)
>>   {
>>       int ret;
>> -    static int once;
>>         ret = kvm_loongarch_put_regs_core(cs);
>>       if (ret) {
>> @@ -843,14 +842,6 @@ int kvm_arch_put_registers(CPUState *cs, KvmPutState level, Error **errp)
>>           return ret;
>>       }
>>   -    if (!once) {
>> -        ret = kvm_set_pv_features(cs);
>> -        if (ret) {
>> -            return ret;
>> -        }
>> -        once = 1;
>> -    }
>> -
>>       if (level >= KVM_PUT_FULL_STATE) {
>>           /*
>>            * only KVM_PUT_FULL_STATE is required, kvm kernel will clear
>> @@ -1209,6 +1200,12 @@ int kvm_arch_init_vcpu(CPUState *cs)
>>           return ret;
>>       }
>>   +    /* pv_features is a per-vCPU attribute; set it here, once per vCPU. */
>> +    ret = kvm_set_pv_features(cs);
>> +    if (ret < 0) {
>> +        return ret;
>> +    }
> I prefer the old method. The pv_features is one vCPU state instead, and it will be added in vCPU VMState if migration is supported on different host kernel version.
> 
> Here is host feature detection, not vCPU state setting.

Thanks, that makes sense. init_vcpu() is the host-detection path; the
per-vCPU write belongs in kvm_arch_put_registers(). I'll keep it there,
drop the VM-global static guard, and push per-vCPU under
level >= KVM_PUT_FULL_STATE, the same gate kvm_set_stealtime() already
uses just below.

For migration across kernel versions, I'll add env->pv_features as a
sub-section of vmstate_loongarch_cpu (leaving version 4 untouched) so
the value travels with the vCPU.

Thanks,
Tao

> 
> Regards
> Bibo Mao
>> +
>>       ret = kvm_cpu_check_ptw(cs, &local_err);
>>       if (ret < 0) {
>>           error_report_err(local_err);
>>
>