Re: [RFC PATCH] kvm/riscv: Add ctxsstatus and ctxhstatus for migration

From: "Radim Krčmář" <rkrcmar@ventanamicro.com>
To: "Jinyu Tang" <tjytimi@163.com>,
	"Anup Patel" <anup@brainfault.org>,
	"Atish Patra" <atish.patra@linux.dev>,
	"Andrew Jones" <ajones@ventanamicro.com>,
	"Conor Dooley" <conor.dooley@microchip.com>,
	"Yong-Xuan Wang" <yongxuan.wang@sifive.com>,
	"Paul Walmsley" <paul.walmsley@sifive.com>,
	"Nutty Liu" <nutty.liu@hotmail.com>
Cc: <kvm@vger.kernel.org>, <kvm-riscv@lists.infradead.org>,
	<linux-riscv@lists.infradead.org>, <linux-kernel@vger.kernel.org>,
	"Tianshun Sun" <stsmail163@163.com>
Subject: Re: [RFC PATCH] kvm/riscv: Add ctxsstatus and ctxhstatus for migration
Date: Mon, 15 Sep 2025 18:52:56 +0200	[thread overview]
Message-ID: <DCTJ3N8W4PCL.9XCHFVGS62SF@ventanamicro.com> (raw)
In-Reply-To: <20250915152731.1371067-1-tjytimi@163.com>

2025-09-15T23:27:31+08:00, Jinyu Tang <tjytimi@163.com>:
> When migrating a VM which guest running in user mode 
> (e.g., executing a while(1) application), the target 
> VM fails to run because it loses the information of 
> guest_context.hstatus and guest_context.sstatus. The 
> VM uses the initialized values instead of the correct ones.
>
> This patch adds two new context registers (ctxsstatus and 
> ctxhstatus) to the kvm_vcpu_csr structure and implements 
> the necessary KVM get and set logic to preserve these values 
> during migration.
>
> QEMU needs to be updated to support these new registers.
> See https://github.com/tjy-zhu/qemu
> for the corresponding QEMU changes.
>
> I'm not sure if adding these CSR registers is a right way. RISCV
> KVM doesn't have API to save these two context csrs now. I will
> submit the corresponding QEMU patch to the QEMU community if
> KVM has API to get and set them.

I don't think it is...

> Signed-off-by: Jinyu Tang <tjytimi@163.com>
> Tested-by: Tianshun Sun <stsmail163@163.com>
> ---
> diff --git a/arch/riscv/kvm/vcpu_onereg.c b/arch/riscv/kvm/vcpu_onereg.c
> @@ -489,6 +489,12 @@ static int kvm_riscv_vcpu_general_get_csr(struct kvm_vcpu *vcpu,
>  	if (reg_num >= sizeof(struct kvm_riscv_csr) / sizeof(unsigned long))
>  		return -ENOENT;
>  
> +	if (reg_num == KVM_REG_RISCV_CSR_REG(ctxsstatus))
> +		csr->ctxsstatus = vcpu->arch.guest_context.sstatus;

Userspace can set guest privilege mode via KVM_REG_RISCV_CORE_REG(mode).

This does propagate to guest_context.sstatus, which is otherwise
internal KVM state, so we definitely shouldn't expose it directly.

> +
> +	if (reg_num == KVM_REG_RISCV_CSR_REG(ctxhstatus))
> +		csr->ctxhstatus = vcpu->arch.guest_context.hstatus;

Neither should userspace be able to directly set hstatus.
KVM should derive it from other userspace configuration.

What isn't correctly reflected in hstatus?

Thanks.

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv