Re: [PATCH] LoongArch: KVM: Set max used FPU type with FPU exception

[Bibo Mao]

another more proactive method is to enabled FPU type with max VM 
supported, rather than ever used.

It is more direct and simpler, FPU none --> LASX, no middle state FPU 
--> LASX or LSX --> LASX.

Regards
Bibo Mao

On 2026/3/23 上午11:40, Bibo Mao wrote:
> With FPU save and restore flow, the cost is the same with different
> FPU width 8/16/32 bytes, whatever from CPU cycle and cache line impaction.
>
> Here is to enable FPU with max used type, for example if application
> ever uses LASX instrction, enable FPU with LASX type even with FPU
> exeception. So it can avoid possible LSX/LASX exception in future.
>
> With context switch microbench which may touch FPU and LASX, there is 9%
> improvement when halt_poll_ns is disabled. The command is
> "./context --test=pipe" and source code located at:
>    https://github.com/bibo-mao/context_switch/blob/main/context.c
>
> Original     Wih patch      improvement
> 75232        82440          9%
>
> Signed-off-by: Bibo Mao <maobibo@xxxxxxxxxxx>
> ---
>   arch/loongarch/include/asm/kvm_host.h |  1 +
>   arch/loongarch/kvm/exit.c             |  6 +++---
>   arch/loongarch/kvm/vcpu.c             | 25 ++++++++++---------------
>   3 files changed, 14 insertions(+), 18 deletions(-)
>
> diff --git a/arch/loongarch/include/asm/kvm_host.h b/arch/loongarch/include/asm/kvm_host.h
> index 19eb5e5c3984..7d739fd5cda6 100644
> --- a/arch/loongarch/include/asm/kvm_host.h
> +++ b/arch/loongarch/include/asm/kvm_host.h
> @@ -203,6 +203,7 @@ struct kvm_vcpu_arch {
>   	/* Which auxiliary state is loaded (KVM_LARCH_*) */
>   	unsigned int aux_inuse;
>   	unsigned int aux_ldtype;
> +	unsigned int aux_used;
>   
>   	/* FPU state */
>   	struct loongarch_fpu fpu FPU_ALIGN;
> diff --git a/arch/loongarch/kvm/exit.c b/arch/loongarch/kvm/exit.c
> index da0ad89f2eb7..da8e330c8b20 100644
> --- a/arch/loongarch/kvm/exit.c
> +++ b/arch/loongarch/kvm/exit.c
> @@ -754,7 +754,7 @@ static int kvm_handle_fpu_disabled(struct kvm_vcpu *vcpu, int ecode)
>   		return RESUME_HOST;
>   	}
>   
> -	vcpu->arch.aux_ldtype = KVM_LARCH_FPU;
> +	vcpu->arch.aux_ldtype = vcpu->arch.aux_used | KVM_LARCH_FPU;
>   	kvm_make_request(KVM_REQ_AUX_LOAD, vcpu);
>   
>   	return RESUME_GUEST;
> @@ -796,7 +796,7 @@ static int kvm_handle_lsx_disabled(struct kvm_vcpu *vcpu, int ecode)
>   	if (!kvm_guest_has_lsx(&vcpu->arch))
>   		kvm_queue_exception(vcpu, EXCCODE_INE, 0);
>   	else {
> -		vcpu->arch.aux_ldtype = KVM_LARCH_LSX;
> +		vcpu->arch.aux_ldtype = vcpu->arch.aux_used | KVM_LARCH_LSX;
>   		kvm_make_request(KVM_REQ_AUX_LOAD, vcpu);
>   	}
>   
> @@ -816,7 +816,7 @@ static int kvm_handle_lasx_disabled(struct kvm_vcpu *vcpu, int ecode)
>   	if (!kvm_guest_has_lasx(&vcpu->arch))
>   		kvm_queue_exception(vcpu, EXCCODE_INE, 0);
>   	else {
> -		vcpu->arch.aux_ldtype = KVM_LARCH_LASX;
> +		vcpu->arch.aux_ldtype = vcpu->arch.aux_used | KVM_LARCH_LASX;
>   		kvm_make_request(KVM_REQ_AUX_LOAD, vcpu);
>   	}
>   
> diff --git a/arch/loongarch/kvm/vcpu.c b/arch/loongarch/kvm/vcpu.c
> index 8810fcd7e26e..0a111c74a330 100644
> --- a/arch/loongarch/kvm/vcpu.c
> +++ b/arch/loongarch/kvm/vcpu.c
> @@ -239,23 +239,14 @@ static void kvm_late_check_requests(struct kvm_vcpu *vcpu)
>   		}
>   
>   	if (kvm_check_request(KVM_REQ_AUX_LOAD, vcpu)) {
> -		switch (vcpu->arch.aux_ldtype) {
> -		case KVM_LARCH_FPU:
> -			kvm_own_fpu(vcpu);
> -			break;
> -		case KVM_LARCH_LSX:
> -			kvm_own_lsx(vcpu);
> -			break;
> -		case KVM_LARCH_LASX:
> +		if (vcpu->arch.aux_ldtype & KVM_LARCH_LASX)
>   			kvm_own_lasx(vcpu);
> -			break;
> -		case KVM_LARCH_LBT:
> +		else if (vcpu->arch.aux_ldtype & KVM_LARCH_LSX)
> +			kvm_own_lsx(vcpu);
> +		else if (vcpu->arch.aux_ldtype & KVM_LARCH_FPU)
> +			kvm_own_fpu(vcpu);
> +		else if (vcpu->arch.aux_ldtype == KVM_LARCH_LBT)
>   			kvm_own_lbt(vcpu);
> -			break;
> -		default:
> -			break;
> -		}
> -
>   		vcpu->arch.aux_ldtype = 0;
>   	}
>   }
> @@ -956,6 +947,7 @@ static int kvm_set_one_reg(struct kvm_vcpu *vcpu,
>   			break;
>   		case KVM_REG_LOONGARCH_VCPU_RESET:
>   			vcpu->arch.st.guest_addr = 0;
> +			vcpu->arch.aux_used = 0;
>   			memset(&vcpu->arch.irq_pending, 0, sizeof(vcpu->arch.irq_pending));
>   			memset(&vcpu->arch.irq_clear, 0, sizeof(vcpu->arch.irq_clear));
>   
> @@ -1384,6 +1376,7 @@ void kvm_own_fpu(struct kvm_vcpu *vcpu)
>   
>   	kvm_restore_fpu(&vcpu->arch.fpu);
>   	vcpu->arch.aux_inuse |= KVM_LARCH_FPU;
> +	vcpu->arch.aux_used |= KVM_LARCH_FPU;
>   	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_FPU);
>   }
>   
> @@ -1412,6 +1405,7 @@ int kvm_own_lsx(struct kvm_vcpu *vcpu)
>   
>   	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_LSX);
>   	vcpu->arch.aux_inuse |= KVM_LARCH_LSX | KVM_LARCH_FPU;
> +	vcpu->arch.aux_used |= KVM_LARCH_LSX | KVM_LARCH_FPU;
>   
>   	return 0;
>   }
> @@ -1442,6 +1436,7 @@ int kvm_own_lasx(struct kvm_vcpu *vcpu)
>   
>   	trace_kvm_aux(vcpu, KVM_TRACE_AUX_RESTORE, KVM_TRACE_AUX_LASX);
>   	vcpu->arch.aux_inuse |= KVM_LARCH_LASX | KVM_LARCH_LSX | KVM_LARCH_FPU;
> +	vcpu->arch.aux_used |= KVM_LARCH_LASX | KVM_LARCH_LSX | KVM_LARCH_FPU;
>   
>   	return 0;
>   }
>
> base-commit: c369299895a591d96745d6492d4888259b004a9e