[v2] target/loongarch: fix vcpu reset command word issue

[PATCH V2] target/loongarch: fix vcpu reset command word issue

Posted by Xianglai Li 12 months ago

When the KVM_REG_LOONGARCH_VCPU_RESET command word
is sent to the kernel through the kvm_set_one_reg interface,
the parameter source needs to be a legal address,
otherwise the kernel will return an error and the command word
will fail to be sent.

Signed-off-by: Xianglai Li <lixianglai@loongson.cn>
---
Cc: Bibo Mao <Maobibo@loongson.cn>
Cc: Philippe Mathieu-Daudé <philmd@linaro.org>
Cc: Song Gao <gaosong@loongson.cn>
Cc: Xianglai Li <lixianglai@loongson.cn>

CHANGE:
V2<-V1:
  1.Sets the initial value of the variable and
  adds a function return value judgment and prints a log

 target/loongarch/kvm/kvm.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/target/loongarch/kvm/kvm.c b/target/loongarch/kvm/kvm.c
index a3f55155b0..3f499e60ab 100644
--- a/target/loongarch/kvm/kvm.c
+++ b/target/loongarch/kvm/kvm.c
@@ -581,9 +581,14 @@ static int kvm_loongarch_get_lbt(CPUState *cs)
 void kvm_arch_reset_vcpu(CPUState *cs)
 {
     CPULoongArchState *env = cpu_env(cs);
+    int ret = 0;
+    uint64_t unused = 0;
 
     env->mp_state = KVM_MP_STATE_RUNNABLE;
-    kvm_set_one_reg(cs, KVM_REG_LOONGARCH_VCPU_RESET, 0);
+    ret = kvm_set_one_reg(cs, KVM_REG_LOONGARCH_VCPU_RESET, &unused);
+    if (ret) {
+        error_report("Failed to set KVM_REG_LOONGARCH_VCPU_RESET");
+    }
 }
 
 static int kvm_loongarch_get_mpstate(CPUState *cs)
-- 
2.39.1

Re: [PATCH V2] target/loongarch: fix vcpu reset command word issue

Posted by Philippe Mathieu-Daudé 12 months ago

Hi,

On 8/2/25 08:50, Xianglai Li wrote:
> When the KVM_REG_LOONGARCH_VCPU_RESET command word
> is sent to the kernel through the kvm_set_one_reg interface,
> the parameter source needs to be a legal address,
> otherwise the kernel will return an error and the command word
> will fail to be sent.
> 
> Signed-off-by: Xianglai Li <lixianglai@loongson.cn>
> ---
> Cc: Bibo Mao <Maobibo@loongson.cn>
> Cc: Philippe Mathieu-Daudé <philmd@linaro.org>
> Cc: Song Gao <gaosong@loongson.cn>
> Cc: Xianglai Li <lixianglai@loongson.cn>
> 
> CHANGE:
> V2<-V1:
>    1.Sets the initial value of the variable and
>    adds a function return value judgment and prints a log
> 
>   target/loongarch/kvm/kvm.c | 7 ++++++-
>   1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/target/loongarch/kvm/kvm.c b/target/loongarch/kvm/kvm.c
> index a3f55155b0..3f499e60ab 100644
> --- a/target/loongarch/kvm/kvm.c
> +++ b/target/loongarch/kvm/kvm.c
> @@ -581,9 +581,14 @@ static int kvm_loongarch_get_lbt(CPUState *cs)
>   void kvm_arch_reset_vcpu(CPUState *cs)
>   {
>       CPULoongArchState *env = cpu_env(cs);
> +    int ret = 0;
> +    uint64_t unused = 0;
>   
>       env->mp_state = KVM_MP_STATE_RUNNABLE;
> -    kvm_set_one_reg(cs, KVM_REG_LOONGARCH_VCPU_RESET, 0);
> +    ret = kvm_set_one_reg(cs, KVM_REG_LOONGARCH_VCPU_RESET, &unused);
> +    if (ret) {
> +        error_report("Failed to set KVM_REG_LOONGARCH_VCPU_RESET");

If this call fails, I'd not rely on the state of the VM. What about:

if (ret < 0) {
     error_report("Failed to set KVM_REG_LOONGARCH_VCPU_RESET: %s",
                  strerror(errno));
     exit(EXIT_FAILURE);
}

?

> +    }
>   }
>   
>   static int kvm_loongarch_get_mpstate(CPUState *cs)

Re: [PATCH V2] target/loongarch: fix vcpu reset command word issue

Posted by lixianglai 12 months ago

Hi Philippe Mathieu-Daudé:
> Hi,
>
> On 8/2/25 08:50, Xianglai Li wrote:
>> When the KVM_REG_LOONGARCH_VCPU_RESET command word
>> is sent to the kernel through the kvm_set_one_reg interface,
>> the parameter source needs to be a legal address,
>> otherwise the kernel will return an error and the command word
>> will fail to be sent.
>>
>> Signed-off-by: Xianglai Li <lixianglai@loongson.cn>
>> ---
>> Cc: Bibo Mao <Maobibo@loongson.cn>
>> Cc: Philippe Mathieu-Daudé <philmd@linaro.org>
>> Cc: Song Gao <gaosong@loongson.cn>
>> Cc: Xianglai Li <lixianglai@loongson.cn>
>>
>> CHANGE:
>> V2<-V1:
>>    1.Sets the initial value of the variable and
>>    adds a function return value judgment and prints a log
>>
>>   target/loongarch/kvm/kvm.c | 7 ++++++-
>>   1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/target/loongarch/kvm/kvm.c b/target/loongarch/kvm/kvm.c
>> index a3f55155b0..3f499e60ab 100644
>> --- a/target/loongarch/kvm/kvm.c
>> +++ b/target/loongarch/kvm/kvm.c
>> @@ -581,9 +581,14 @@ static int kvm_loongarch_get_lbt(CPUState *cs)
>>   void kvm_arch_reset_vcpu(CPUState *cs)
>>   {
>>       CPULoongArchState *env = cpu_env(cs);
>> +    int ret = 0;
>> +    uint64_t unused = 0;
>>         env->mp_state = KVM_MP_STATE_RUNNABLE;
>> -    kvm_set_one_reg(cs, KVM_REG_LOONGARCH_VCPU_RESET, 0);
>> +    ret = kvm_set_one_reg(cs, KVM_REG_LOONGARCH_VCPU_RESET, &unused);
>> +    if (ret) {
>> +        error_report("Failed to set KVM_REG_LOONGARCH_VCPU_RESET");
>
> If this call fails, I'd not rely on the state of the VM. What about:
>
> if (ret < 0) {
>     error_report("Failed to set KVM_REG_LOONGARCH_VCPU_RESET: %s",
>                  strerror(errno));
>     exit(EXIT_FAILURE);
> }
>
> ?
>
I'll second that!

Thanks,
Xianglai.
>> +    }
>>   }
>>     static int kvm_loongarch_get_mpstate(CPUState *cs)
>