Re: [PATCH v4 2/2] s390/kvm: diagnose 318 handling

[David Hildenbrand <david@redhat.com>]

On 02.05.19 00:51, Collin Walling wrote:
> DIAGNOSE 0x318 (diag318) is a privileged s390x instruction that must
> be intercepted by SIE and handled via KVM. Let's introduce some
> functions to communicate between userspace and KVM via ioctls. These
> will be used to get/set the diag318 related information (also known
> as the "Control Program Code" or "CPC"), as well as check the system
> if KVM supports handling this instruction.
> 
> This information can help with diagnosing the OS the VM is running
> in (Linux, z/VM, etc) if the OS calls this instruction.
> 
> The get/set functions are introduced primarily for VM migration and
> reset, though no harm could be done to the system if a userspace
> program decides to alter this data (this is highly discouraged).
> 
> The Control Program Name Code (CPNC) is stored in the SIE block and
> a copy is retained in each VCPU. The Control Program Version Code
> (CPVC) retains a copy in each VCPU as well.
> 
> At this time, the CPVC is not reported as its format is yet to be
> defined.
> 
> Note that the CPNC is set in the SIE block iff the host hardware
> supports it.

For vSIE and SIE you only configure the CPNC. Is that sufficient?
Shouldn't diag318 allow the guest to set both? (especially regarding vSIE)

[...]
> 
> diff --git a/Documentation/virtual/kvm/devices/vm.txt b/Documentation/virtual/kvm/devices/vm.txt
> index 95ca68d..9a8d934 100644
> --- a/Documentation/virtual/kvm/devices/vm.txt
> +++ b/Documentation/virtual/kvm/devices/vm.txt
> @@ -267,3 +267,17 @@ Parameters: address of a buffer in user space to store the data (u64) to;
>  	    if it is enabled
>  Returns:    -EFAULT if the given address is not accessible from kernel space
>  	    0 in case of success.
> +
> +6. GROUP: KVM_S390_VM_MISC
> +Architectures: s390
> +
> +6.1. KVM_S390_VM_MISC_CPC (r/w)
> +
> +Allows userspace to access the "Control Program Code" which consists of a
> +1-byte "Control Program Name Code" and a 7-byte "Control Program Version Code".
> +This information is initialized during IPL and must be preserved during
> +migration.

Your implementation does not match this description. User space can only
get/set the cpnc effectively for the HW to see it, not the CPVC, no?

Shouldn't you transparently forward that data to the SCB for vSIE/SIE,
because we really don't care what the target format will be?

> +
> +Parameters: address of a buffer in user space to store the data (u64) to
> +Returns:    -EFAULT if the given address is not accessible from kernel space
> +	     0 in case of success.

[...]
>  
>  #define KVM_HVA_ERR_BAD		(-1UL)
> diff --git a/arch/s390/include/uapi/asm/kvm.h b/arch/s390/include/uapi/asm/kvm.h
> index 16511d9..3d3d2a5 100644
> --- a/arch/s390/include/uapi/asm/kvm.h
> +++ b/arch/s390/include/uapi/asm/kvm.h
> @@ -74,6 +74,7 @@ struct kvm_s390_io_adapter_req {
>  #define KVM_S390_VM_CRYPTO		2
>  #define KVM_S390_VM_CPU_MODEL		3
>  #define KVM_S390_VM_MIGRATION		4
> +#define KVM_S390_VM_MISC		5
>  
>  /* kvm attributes for mem_ctrl */
>  #define KVM_S390_VM_MEM_ENABLE_CMMA	0
> @@ -168,6 +169,9 @@ struct kvm_s390_vm_cpu_subfunc {
>  #define KVM_S390_VM_MIGRATION_START	1
>  #define KVM_S390_VM_MIGRATION_STATUS	2
>  
> +/* kvm attributes for KVM_S390_VM_MISC */
> +#define KVM_S390_VM_MISC_CPC		0
> +
>  /* for KVM_GET_REGS and KVM_SET_REGS */
>  struct kvm_regs {
>  	/* general purpose regs for s390 */
> diff --git a/arch/s390/kvm/diag.c b/arch/s390/kvm/diag.c
> index 45634b3d..9762e6a 100644
> --- a/arch/s390/kvm/diag.c
> +++ b/arch/s390/kvm/diag.c
> @@ -235,6 +235,21 @@ static int __diag_virtio_hypercall(struct kvm_vcpu *vcpu)
>  	return ret < 0 ? ret : 0;
>  }
>  
> +static int __diag_set_control_prog_name(struct kvm_vcpu *vcpu)

Can we name that "__diag_set_cpc" ?

"control_prog_name" is certainly not 100% correct.

> +{
> +	unsigned int reg = (vcpu->arch.sie_block->ipa & 0xf0) >> 4;
> +	u64 cpc = vcpu->run->s.regs.gprs[reg];
> +
> +	vcpu->stat.diagnose_318++;
> +	kvm_s390_set_cpc(vcpu->kvm, cpc);
> +
> +	VCPU_EVENT(vcpu, 3, "diag 0x318 cpnc: 0x%x cpvc: 0x%llx",
> +		   vcpu->kvm->arch.diag318_info.cpnc,
> +		   (u64)vcpu->kvm->arch.diag318_info.cpvc);
> +
> +	return 0;
> +}


[...]
> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
> index 4638303..910af18 100644
> --- a/arch/s390/kvm/kvm-s390.c
> +++ b/arch/s390/kvm/kvm-s390.c
> @@ -156,6 +156,7 @@ struct kvm_stats_debugfs_item debugfs_entries[] = {
>  	{ "instruction_diag_9c", VCPU_STAT(diagnose_9c) },
>  	{ "instruction_diag_258", VCPU_STAT(diagnose_258) },
>  	{ "instruction_diag_308", VCPU_STAT(diagnose_308) },
> +	{ "instruction_diag_318", VCPU_STAT(diagnose_318) },
>  	{ "instruction_diag_500", VCPU_STAT(diagnose_500) },
>  	{ "instruction_diag_other", VCPU_STAT(diagnose_other) },
>  	{ NULL }
> @@ -1190,6 +1191,70 @@ static int kvm_s390_get_tod(struct kvm *kvm, struct kvm_device_attr *attr)
>  	return ret;
>  }
>  
> +void kvm_s390_set_cpc(struct kvm *kvm, u64 cpc)
> +{
> +	struct kvm_vcpu *vcpu;
> +	int i;
> +
> +	mutex_lock(&kvm->lock);
> +	kvm->arch.diag318_info.val = cpc;
> +
> +	VM_EVENT(kvm, 3, "SET: CPNC: 0x%x CPVC: 0x%llx",
> +		 kvm->arch.diag318_info.cpnc, (u64)kvm->arch.diag318_info.cpvc);
> +
> +	if (sclp.has_diag318) {
> +		kvm_for_each_vcpu(i, vcpu, kvm) {
> +			vcpu->arch.sie_block->cpnc = kvm->arch.diag318_info.cpnc;
> +		}
> +	}

Do we care about races here between guest VCPUs reading it via the SCB
(HW) and us changing the value? My gut feeling is that it can be tolerated.

> +	mutex_unlock(&kvm->lock);
> +}
> +
> +static int kvm_s390_set_misc(struct kvm *kvm, struct kvm_device_attr *attr)
> +{
> +	int ret;
> +	u64 cpc;
> +
> +	switch (attr->attr) {
> +	case KVM_S390_VM_MISC_CPC:
> +		ret = -EFAULT;
> +		if (get_user(cpc, (u64 __user *)attr->addr))
> +			break;
> +		kvm_s390_set_cpc(kvm, cpc);
> +		ret = 0;
> +		break;
> +	default:
> +		ret = -ENXIO;
> +		break;
> +	}
> +	return ret;
> +}
> +
> +static int kvm_s390_get_cpc(struct kvm *kvm, struct kvm_device_attr *attr)
> +{
> +	if (put_user(kvm->arch.diag318_info.val, (u64 __user *)attr->addr))
> +		return -EFAULT;

Another possible race with setting code. Should be at least take the
kvm->lock here? Otherwise, also looks like this can be tolerated.

-- 

Thanks,

David / dhildenb