Received-SPF: pass (google.com: domain of linux-kernel+bounces-204207-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Subject: Re: [PATCH] LoongArch: KVM: Add feature passing from user space
To: WANG Xuerui <kernel@xen0n.name>, Tianrui Zhao <zhaotianrui@loongson.cn>,
 Huacai Chen <chenhuacai@kernel.org>
Cc: kvm@vger.kernel.org, loongarch@lists.linux.dev,
 linux-kernel@vger.kernel.org
References: <20240606074850.2651896-1-maobibo@loongson.cn>
 <9bb552c8-fe86-43dc-9c4e-0b95c99fb25c@xen0n.name>
From: maobibo <maobibo@loongson.cn>
Message-ID: <2774b010-8033-2167-474a-cb1b29b27d2b@loongson.cn>
Date: Thu, 6 Jun 2024 19:54:58 +0800
User-Agent: Mozilla/5.0 (X11; Linux loongarch64; rv:68.0) Gecko/20100101
 Thunderbird/68.7.0
Precedence: bulk
MIME-Version: 1.0
In-Reply-To: <9bb552c8-fe86-43dc-9c4e-0b95c99fb25c@xen0n.name>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit

Xuerui,

Thanks for your reviewing.
I reply inline.

On 2024/6/6 下午7:20, WANG Xuerui wrote:
> Hi,
> 
> On 6/6/24 15:48, Bibo Mao wrote:
>> Currently features defined in cpucfg CPUCFG_KVM_FEATURE comes from
>> kvm kernel mode only. Some features are defined in user space VMM,
> 
> "come from kernel side only. But currently KVM is not aware of 
> user-space VMM features which makes it hard to employ optimizations that 
> are both (1) specific to the VM use case and (2) requiring cooperation 
> from user space."
Will modify in next version.
> 
>> however KVM module does not know. Here interface is added to update
>> register CPUCFG_KVM_FEATURE from user space, only bit 24 - 31 is valid.
>>
>> Feature KVM_LOONGARCH_VCPU_FEAT_VIRT_EXTIOI is added from user mdoe.
>> FEAT_VIRT_EXTIOI is virt EXTIOI extension which can route interrupt
>> to 256 VCPUs rather than 4 CPUs like real hw.
> 
> "A new feature bit ... is added which can be set from user space. 
> FEAT_... indicates that the VM EXTIOI can route interrupts to 256 vCPUs, 
> rather than 4 like with real HW."
will modify in next version.

> 
> (Am I right in paraphrasing the "EXTIOI" part?)
> 
>>
>> Signed-off-by: Bibo Mao <maobibo@loongson.cn>
>> ---
>>   arch/loongarch/include/asm/kvm_host.h  |  4 +++
>>   arch/loongarch/include/asm/loongarch.h |  5 ++++
>>   arch/loongarch/include/uapi/asm/kvm.h  |  2 ++
>>   arch/loongarch/kvm/exit.c              |  1 +
>>   arch/loongarch/kvm/vcpu.c              | 36 +++++++++++++++++++++++---
>>   5 files changed, 44 insertions(+), 4 deletions(-)
>>
>> diff --git a/arch/loongarch/include/asm/kvm_host.h 
>> b/arch/loongarch/include/asm/kvm_host.h
>> index 88023ab59486..8fa50d757247 100644
>> --- a/arch/loongarch/include/asm/kvm_host.h
>> +++ b/arch/loongarch/include/asm/kvm_host.h
>> @@ -135,6 +135,9 @@ enum emulation_result {
>>   #define KVM_LARCH_HWCSR_USABLE    (0x1 << 4)
>>   #define KVM_LARCH_LBT        (0x1 << 5)
>> +#define KVM_LOONGARCH_USR_FEAT_MASK            \
>> +    BIT(KVM_LOONGARCH_VCPU_FEAT_VIRT_EXTIOI)
>> +
>>   struct kvm_vcpu_arch {
>>       /*
>>        * Switch pointer-to-function type to unsigned long
>> @@ -210,6 +213,7 @@ struct kvm_vcpu_arch {
>>           u64 last_steal;
>>           struct gfn_to_hva_cache cache;
>>       } st;
>> +    unsigned int usr_features;
>>   };
>>   static inline unsigned long readl_sw_gcsr(struct loongarch_csrs 
>> *csr, int reg)
>> diff --git a/arch/loongarch/include/asm/loongarch.h 
>> b/arch/loongarch/include/asm/loongarch.h
>> index 7a4633ef284b..4d9837512c19 100644
>> --- a/arch/loongarch/include/asm/loongarch.h
>> +++ b/arch/loongarch/include/asm/loongarch.h
>> @@ -167,9 +167,14 @@
>>   #define CPUCFG_KVM_SIG            (CPUCFG_KVM_BASE + 0)
>>   #define  KVM_SIGNATURE            "KVM\0"
>> +/*
>> + * BIT 24 - 31 is features configurable by user space vmm
>> + */
>>   #define CPUCFG_KVM_FEATURE        (CPUCFG_KVM_BASE + 4)
>>   #define  KVM_FEATURE_IPI        BIT(1)
>>   #define  KVM_FEATURE_STEAL_TIME        BIT(2)
>> +/* With VIRT_EXTIOI feature, interrupt can route to 256 VCPUs */
>> +#define  KVM_FEATURE_VIRT_EXTIOI    BIT(24)
>>   #ifndef __ASSEMBLY__
> 
> What about assigning a new CPUCFG leaf ID for separating the two kinds 
> of feature flags very cleanly?
For compatible issue like new kernel on old KVM host, to add a new
CPUCFG leaf ID, a new feature need be defined on existing 
CPUCFG_KVM_FEATURE register. Such as:
    #define  KVM_FEATURE_EXTEND_CPUCFG        BIT(3)

VM need check feature KVM_FEATURE_EXTEND_CPUCFG at first, and then read 
the new CPUCFG leaf ID if feature EXTEND_CPUCFG is enabled.

That maybe makes it complicated since feature bit is enough now.
> 
>> @@ -896,7 +907,24 @@ static int kvm_loongarch_vcpu_get_attr(struct 
>> kvm_vcpu *vcpu,
>>   static int kvm_loongarch_cpucfg_set_attr(struct kvm_vcpu *vcpu,
>>                        struct kvm_device_attr *attr)
>>   {
>> -    return -ENXIO;
>> +    u64 __user *user = (u64 __user *)attr->addr;
>> +    u64 val, valid_flags;
>> +
>> +    switch (attr->attr) {
>> +    case CPUCFG_KVM_FEATURE:
>> +        if (get_user(val, user))
>> +            return -EFAULT;
>> +
>> +        valid_flags = KVM_LOONGARCH_USR_FEAT_MASK;
>> +        if (val & ~valid_flags)
>> +            return -EINVAL;
>> +
>> +        vcpu->arch.usr_features |= val;
> 
> Isn't this usage of "|=" instead of "=" implying that the feature bits 
> could not get disabled after being enabled earlier, for whatever reason?
yes, "=" is better. Will modify in next version.

Regards
Bibo Mao
> 
>> +        return 0;
>> +
>> +    default:
>> +        return -ENXIO;
>> +    }
>>   }
>>   static int kvm_loongarch_vcpu_set_attr(struct kvm_vcpu *vcpu,
>>
>> base-commit: 2df0193e62cf887f373995fb8a91068562784adc
>