Received-SPF: pass (google.com: domain of linux-kernel+bounces-54653-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
From: Michael Ellerman <mpe@ellerman.id.au>
To: Amit Machhiwal <amachhiw@linux.ibm.com>, linuxppc-dev@lists.ozlabs.org,
 kvm@vger.kernel.org, kvm-ppc@vger.kernel.org
Cc: Vaibhav Jain <vaibhav@linux.ibm.com>, Nicholas Piggin
 <npiggin@gmail.com>, Jordan Niethe <jniethe5@gmail.com>, Vaidyanathan
 Srinivasan <svaidy@linux.ibm.com>, "Aneesh Kumar K . V"
 <aneesh.kumar@kernel.org>, "Naveen N . Rao" <naveen.n.rao@linux.ibm.com>,
 Christophe Leroy <christophe.leroy@csgroup.eu>, Amit Machhiwal
 <amachhiw@linux.ibm.com>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v3] KVM: PPC: Book3S HV: Fix L2 guest reboot failure due
 to empty 'arch_compat'
In-Reply-To: <20240205181833.212955-1-amachhiw@linux.ibm.com>
References: <20240205181833.212955-1-amachhiw@linux.ibm.com>
Date: Tue, 06 Feb 2024 21:09:48 +1100
Message-ID: <87r0hp9a4z.fsf@mail.lhotse>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain

Hi Amit,

One comment below ...

Amit Machhiwal <amachhiw@linux.ibm.com> writes:
> Currently, rebooting a pseries nested qemu-kvm guest (L2) results in
> below error as L1 qemu sends PVR value 'arch_compat' == 0 via
> ppc_set_compat ioctl. This triggers a condition failure in
> kvmppc_set_arch_compat() resulting in an EINVAL.
..
>  	
> diff --git a/arch/powerpc/kvm/book3s_hv_nestedv2.c b/arch/powerpc/kvm/book3s_hv_nestedv2.c
> index 5378eb40b162..6042bdc70230 100644
> --- a/arch/powerpc/kvm/book3s_hv_nestedv2.c
> +++ b/arch/powerpc/kvm/book3s_hv_nestedv2.c
> @@ -347,8 +348,26 @@ static int gs_msg_ops_vcpu_fill_info(struct kvmppc_gs_buff *gsb,
>  			break;
>  		}
>  		case KVMPPC_GSID_LOGICAL_PVR:
> -			rc = kvmppc_gse_put_u32(gsb, iden,
> -						vcpu->arch.vcore->arch_compat);
> +			/*
> +			 * Though 'arch_compat == 0' would mean the default
> +			 * compatibility, arch_compat, being a Guest Wide
> +			 * Element, cannot be filled with a value of 0 in GSB
> +			 * as this would result into a kernel trap.
> +			 * Hence, when `arch_compat == 0`, arch_compat should
> +			 * default to L1's PVR.
> +			 *
> +			 * Rework this when PowerVM supports a value of 0
> +			 * for arch_compat for KVM API v2.
> +			 */

Is there an actual plan that PowerVM will support this in future?

If so, how will a future kernel know that it's running on a version of
PowerVM that does support arch_compat == 0?

Similarly how will we know when it's OK to drop support for this
workaround?

cheers