Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Cornelia Huck <cohuck@redhat.com>
To:     David Hildenbrand <david@redhat.com>,
        Pierre Morel <pmorel@linux.ibm.com>, kvm@vger.kernel.org
Cc:     linux-s390@vger.kernel.org, linux-kernel@vger.kernel.org,
        borntraeger@de.ibm.com, frankja@linux.ibm.com, thuth@redhat.com,
        imbrenda@linux.ibm.com, hca@linux.ibm.com, gor@linux.ibm.com
Subject: Re: [PATCH v1 1/2] s390x: KVM: accept STSI for CPU topology
 information
In-Reply-To: <57e57ba5-62ea-f1ff-0d83-5605d57be92d@redhat.com>
Organization: Red Hat GmbH
References: <1626276343-22805-1-git-send-email-pmorel@linux.ibm.com>
 <1626276343-22805-2-git-send-email-pmorel@linux.ibm.com>
 <db788a8c-99a9-6d99-07ab-b49e953d91a2@redhat.com>
 <87fswfdiuu.fsf@redhat.com>
 <57e57ba5-62ea-f1ff-0d83-5605d57be92d@redhat.com>
User-Agent: Notmuch/0.32.1 (https://notmuchmail.org)
Date:   Thu, 15 Jul 2021 12:16:06 +0200
Message-ID: <87czrjdgrd.fsf@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

On Thu, Jul 15 2021, David Hildenbrand <david@redhat.com> wrote:

> On 15.07.21 11:30, Cornelia Huck wrote:
>> On Thu, Jul 15 2021, David Hildenbrand <david@redhat.com> wrote:
>> 
>>> On 14.07.21 17:25, Pierre Morel wrote:
>>>> STSI(15.1.x) gives information on the CPU configuration topology.
>>>> Let's accept the interception of STSI with the function code 15 and
>>>> let the userland part of the hypervisor handle it.
>>>>
>>>> Signed-off-by: Pierre Morel <pmorel@linux.ibm.com>
>>>> ---
>>>>    arch/s390/kvm/priv.c | 11 ++++++++++-
>>>>    1 file changed, 10 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/s390/kvm/priv.c b/arch/s390/kvm/priv.c
>>>> index 9928f785c677..4ab5f8b7780e 100644
>>>> --- a/arch/s390/kvm/priv.c
>>>> +++ b/arch/s390/kvm/priv.c
>>>> @@ -856,7 +856,7 @@ static int handle_stsi(struct kvm_vcpu *vcpu)
>>>>    	if (vcpu->arch.sie_block->gpsw.mask & PSW_MASK_PSTATE)
>>>>    		return kvm_s390_inject_program_int(vcpu, PGM_PRIVILEGED_OP);
>>>>    
>>>> -	if (fc > 3) {
>>>> +	if (fc > 3 && fc != 15) {
>>>>    		kvm_s390_set_psw_cc(vcpu, 3);
>>>>    		return 0;
>>>>    	}
>>>> @@ -893,6 +893,15 @@ static int handle_stsi(struct kvm_vcpu *vcpu)
>>>>    			goto out_no_data;
>>>>    		handle_stsi_3_2_2(vcpu, (void *) mem);
>>>>    		break;
>>>> +	case 15:
>>>> +		if (sel1 != 1 || sel2 < 2 || sel2 > 6)
>>>> +			goto out_no_data;
>>>> +		if (vcpu->kvm->arch.user_stsi) {
>>>> +			insert_stsi_usr_data(vcpu, operand2, ar, fc, sel1, sel2);
>>>> +			return -EREMOTE;
>> 
>> This bypasses the trace event further down.
>> 
>>>> +		}
>>>> +		kvm_s390_set_psw_cc(vcpu, 3);
>>>> +		return 0;
>>>>    	}
>>>>    	if (kvm_s390_pv_cpu_is_protected(vcpu)) {
>>>>    		memcpy((void *)sida_origin(vcpu->arch.sie_block), (void *)mem,
>>> 3. User space awareness
>>>
>>> How can user space identify that we actually forward these intercepts?
>>> How can it enable them? The old KVM_CAP_S390_USER_STSI capability
>>> is not sufficient.
>> 
>> Why do you think that it is not sufficient? USER_STSI basically says
>> "you may get an exit that tells you about a buffer to fill in some more
>> data for a stsi command, and we also tell you which call". If userspace
>> does not know what to add for a certain call, it is free to just do
>> nothing, and if it does not get some calls it would support, that should
>> not be a problem, either?
>
> If you migrate your VM from machine a to machine b, from kernel a to 
> kernel b, and kernel b does not trigger exits to user space for fc=15, 
> how could QEMU spot and catch the different capabilities to make sure 
> the guest can continue using the feature?

Wouldn't that imply that the USER_STSI feature, in the function-agnostic
way it is documented, was broken from the start?

Hm. Maybe we need some kind of facility where userspace can query the
kernel and gets a list of the stsi subcodes it may get exits for, and
possibly fail to start the migration. Having a new capability to be
enabled for every new subcode feels like overkill. I don't think we can
pass a payload ("enable these subfunctions") to a cap.

Or can we tie a subcode to another feature, like the mystery cap that
was mentioned in the description, but does not seem to appear in the
code?