[PATCH] LoongArch: KVM: Access mailbox directly in mail_send()

[PATCH] LoongArch: KVM: Access mailbox directly in mail_send()

[Bibo Mao]

With function mail_send(), it is to write mailbox of other VCPUs.
Existing simple APIs read_mailbox/write_mailbox can be used directly
rather than send command on IOCSR address.

Signed-off-by: Bibo Mao <maobibo@loongson.cn>
---
 arch/loongarch/kvm/intc/ipi.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/arch/loongarch/kvm/intc/ipi.c b/arch/loongarch/kvm/intc/ipi.c
index fe734dc062ed..832b2d4aa2ef 100644
--- a/arch/loongarch/kvm/intc/ipi.c
+++ b/arch/loongarch/kvm/intc/ipi.c
@@ -134,7 +134,8 @@ static int send_ipi_data(struct kvm_vcpu *vcpu, gpa_t addr, uint64_t data)
 
 static int mail_send(struct kvm *kvm, uint64_t data)
 {
-	int cpu, mailbox, offset;
+	int i, cpu, mailbox, offset;
+	uint32_t val = 0, mask = 0;
 	struct kvm_vcpu *vcpu;
 
 	cpu = ((data & 0xffffffff) >> 16) & 0x3ff;
@@ -144,9 +145,18 @@ static int mail_send(struct kvm *kvm, uint64_t data)
 		return -EINVAL;
 	}
 	mailbox = ((data & 0xffffffff) >> 2) & 0x7;
-	offset = IOCSR_IPI_BASE + IOCSR_IPI_BUF_20 + mailbox * 4;
+	offset = IOCSR_IPI_BUF_20 + mailbox * 4;
+	if ((data >> 27) & 0xf) {
+		val = read_mailbox(vcpu, offset, 4);
+		for (i = 0; i < 4; i++)
+			if (data & (BIT(27 + i)))
+				mask |= (0xff << (i * 8));
+		val &= mask;
+	}
 
-	return send_ipi_data(vcpu, offset, data);
+	val |= ((uint32_t)(data >> 32) & ~mask);
+	write_mailbox(vcpu, offset, val, 4);
+	return 0;
 }
 
 static int any_send(struct kvm *kvm, uint64_t data)

base-commit: 038d61fd642278bab63ee8ef722c50d10ab01e8f
-- 
2.39.3

Re: [PATCH] LoongArch: KVM: Access mailbox directly in mail_send()

[Yanteng Si]

在 7/31/25 3:59 PM, Bibo Mao 写道:
> With function mail_send(), it is to write mailbox of other VCPUs.
> Existing simple APIs read_mailbox/write_mailbox can be used directly
> rather than send command on IOCSR address.
Hmm, that's indeed a feasible approach. However, I'm
curious: what is the purpose of designing IOCSR in
LoongArch? Is it merely to make things "complicated"?

Thanks,
Yanteng
> 
> Signed-off-by: Bibo Mao <maobibo@loongson.cn>
> ---
>   arch/loongarch/kvm/intc/ipi.c | 16 +++++++++++++---
>   1 file changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/loongarch/kvm/intc/ipi.c b/arch/loongarch/kvm/intc/ipi.c
> index fe734dc062ed..832b2d4aa2ef 100644
> --- a/arch/loongarch/kvm/intc/ipi.c
> +++ b/arch/loongarch/kvm/intc/ipi.c
> @@ -134,7 +134,8 @@ static int send_ipi_data(struct kvm_vcpu *vcpu, gpa_t addr, uint64_t data)
>   
>   static int mail_send(struct kvm *kvm, uint64_t data)
>   {
> -	int cpu, mailbox, offset;
> +	int i, cpu, mailbox, offset;
> +	uint32_t val = 0, mask = 0;
>   	struct kvm_vcpu *vcpu;
>   
>   	cpu = ((data & 0xffffffff) >> 16) & 0x3ff;
> @@ -144,9 +145,18 @@ static int mail_send(struct kvm *kvm, uint64_t data)
>   		return -EINVAL;
>   	}
>   	mailbox = ((data & 0xffffffff) >> 2) & 0x7;
> -	offset = IOCSR_IPI_BASE + IOCSR_IPI_BUF_20 + mailbox * 4;
> +	offset = IOCSR_IPI_BUF_20 + mailbox * 4;
> +	if ((data >> 27) & 0xf) {
> +		val = read_mailbox(vcpu, offset, 4);
> +		for (i = 0; i < 4; i++)
> +			if (data & (BIT(27 + i)))
> +				mask |= (0xff << (i * 8));
> +		val &= mask;
> +	}
>   
> -	return send_ipi_data(vcpu, offset, data);
> +	val |= ((uint32_t)(data >> 32) & ~mask);
> +	write_mailbox(vcpu, offset, val, 4);
> +	return 0;
>   }
>   
>   static int any_send(struct kvm *kvm, uint64_t data)
> 
> base-commit: 038d61fd642278bab63ee8ef722c50d10ab01e8f

Re: [PATCH] LoongArch: KVM: Access mailbox directly in mail_send()

[Bibo Mao]

On 2025/7/31 下午5:30, Yanteng Si wrote:
> 在 7/31/25 3:59 PM, Bibo Mao 写道:
>> With function mail_send(), it is to write mailbox of other VCPUs.
>> Existing simple APIs read_mailbox/write_mailbox can be used directly
>> rather than send command on IOCSR address.
> Hmm, that's indeed a feasible approach. However, I'm
> curious: what is the purpose of designing IOCSR in
> LoongArch? Is it merely to make things "complicated"?

I am not chip designer -:). MMIO address space is ok from my side, 
however I do not know how to solve multi-chip problem with MMIO method.

Regards
Bibo Mao
> 
> Thanks,
> Yanteng
>>
>> Signed-off-by: Bibo Mao <maobibo@loongson.cn>
>> ---
>>   arch/loongarch/kvm/intc/ipi.c | 16 +++++++++++++---
>>   1 file changed, 13 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/loongarch/kvm/intc/ipi.c 
>> b/arch/loongarch/kvm/intc/ipi.c
>> index fe734dc062ed..832b2d4aa2ef 100644
>> --- a/arch/loongarch/kvm/intc/ipi.c
>> +++ b/arch/loongarch/kvm/intc/ipi.c
>> @@ -134,7 +134,8 @@ static int send_ipi_data(struct kvm_vcpu *vcpu, 
>> gpa_t addr, uint64_t data)
>>   static int mail_send(struct kvm *kvm, uint64_t data)
>>   {
>> -    int cpu, mailbox, offset;
>> +    int i, cpu, mailbox, offset;
>> +    uint32_t val = 0, mask = 0;
>>       struct kvm_vcpu *vcpu;
>>       cpu = ((data & 0xffffffff) >> 16) & 0x3ff;
>> @@ -144,9 +145,18 @@ static int mail_send(struct kvm *kvm, uint64_t data)
>>           return -EINVAL;
>>       }
>>       mailbox = ((data & 0xffffffff) >> 2) & 0x7;
>> -    offset = IOCSR_IPI_BASE + IOCSR_IPI_BUF_20 + mailbox * 4;
>> +    offset = IOCSR_IPI_BUF_20 + mailbox * 4;
>> +    if ((data >> 27) & 0xf) {
>> +        val = read_mailbox(vcpu, offset, 4);
>> +        for (i = 0; i < 4; i++)
>> +            if (data & (BIT(27 + i)))
>> +                mask |= (0xff << (i * 8));
>> +        val &= mask;
>> +    }
>> -    return send_ipi_data(vcpu, offset, data);
>> +    val |= ((uint32_t)(data >> 32) & ~mask);
>> +    write_mailbox(vcpu, offset, val, 4);
>> +    return 0;
>>   }
>>   static int any_send(struct kvm *kvm, uint64_t data)
>>
>> base-commit: 038d61fd642278bab63ee8ef722c50d10ab01e8f
> 

Re: [PATCH] LoongArch: KVM: Access mailbox directly in mail_send()

[Yanteng Si]

在 7/31/25 7:24 PM, Bibo Mao 写道:
> 
> 
> On 2025/7/31 下午5:30, Yanteng Si wrote:
>> 在 7/31/25 3:59 PM, Bibo Mao 写道:
>>> With function mail_send(), it is to write mailbox of other VCPUs.
>>> Existing simple APIs read_mailbox/write_mailbox can be used directly
>>> rather than send command on IOCSR address.
>> Hmm, that's indeed a feasible approach. However, I'm
>> curious: what is the purpose of designing IOCSR in
>> LoongArch? Is it merely to make things "complicated"?
> 
> I am not chip designer -:). MMIO address space is ok from my side, however I do not know how to solve multi-chip problem with MMIO method.

I have reviewed many manuals published on loongson.cn and found
that your chip engineers recommend using IOCSR. Although I don't
know the specific reason, I believe it must have solved some problems,
so let's continue to use IOCSR. see:

《龙芯 3D5000 处理器 寄存器使用手册》第10章

龙芯 3D5000 中支持两种不同的访问方式,一种是地址访问模式,另一种是为了支持处理器寄存器空间的直接指令访问.后面章节进行分别说明.推荐使用处理器 IOCSR 指令访问


Thanks,
Yanteng
> 
> Regards
> Bibo Mao
>>
>> Thanks,
>> Yanteng
>>>
>>> Signed-off-by: Bibo Mao <maobibo@loongson.cn>
>>> ---
>>>   arch/loongarch/kvm/intc/ipi.c | 16 +++++++++++++---
>>>   1 file changed, 13 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/arch/loongarch/kvm/intc/ipi.c b/arch/loongarch/kvm/intc/ipi.c
>>> index fe734dc062ed..832b2d4aa2ef 100644
>>> --- a/arch/loongarch/kvm/intc/ipi.c
>>> +++ b/arch/loongarch/kvm/intc/ipi.c
>>> @@ -134,7 +134,8 @@ static int send_ipi_data(struct kvm_vcpu *vcpu, gpa_t addr, uint64_t data)
>>>   static int mail_send(struct kvm *kvm, uint64_t data)
>>>   {
>>> -    int cpu, mailbox, offset;
>>> +    int i, cpu, mailbox, offset;
>>> +    uint32_t val = 0, mask = 0;
>>>       struct kvm_vcpu *vcpu;
>>>       cpu = ((data & 0xffffffff) >> 16) & 0x3ff;
>>> @@ -144,9 +145,18 @@ static int mail_send(struct kvm *kvm, uint64_t data)
>>>           return -EINVAL;
>>>       }
>>>       mailbox = ((data & 0xffffffff) >> 2) & 0x7;
>>> -    offset = IOCSR_IPI_BASE + IOCSR_IPI_BUF_20 + mailbox * 4;
>>> +    offset = IOCSR_IPI_BUF_20 + mailbox * 4;
>>> +    if ((data >> 27) & 0xf) {
>>> +        val = read_mailbox(vcpu, offset, 4);
>>> +        for (i = 0; i < 4; i++)
>>> +            if (data & (BIT(27 + i)))
>>> +                mask |= (0xff << (i * 8));
>>> +        val &= mask;
>>> +    }
>>> -    return send_ipi_data(vcpu, offset, data);
>>> +    val |= ((uint32_t)(data >> 32) & ~mask);
>>> +    write_mailbox(vcpu, offset, val, 4);
>>> +    return 0;
>>>   }
>>>   static int any_send(struct kvm *kvm, uint64_t data)
>>>
>>> base-commit: 038d61fd642278bab63ee8ef722c50d10ab01e8f
>>
>