Re: [RFC PATCH v2 28/69] KVM: Add per-VM flag to mark read-only memory as unsupported

[Paolo Bonzini]

On 03/07/21 00:04, isaku.yamahata@xxxxxxxxx wrote:
> From: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
>
> Add a flag for TDX to flag RO memory as unsupported and propagate it to
> KVM_MEM_READONLY to allow reporting RO memory as unsupported on a per-VM
> basis.  TDX1 doesn't expose permission bits to the VMM in the SEPT
> tables, i.e. doesn't support read-only private memory.
>
> Signed-off-by: Isaku Yamahata <isaku.yamahata@xxxxxxxxx>
> Co-developed-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> Signed-off-by: Sean Christopherson <sean.j.christopherson@xxxxxxxxx>
> ---
>   arch/x86/kvm/x86.c       | 4 +++-
>   include/linux/kvm_host.h | 4 ++++
>   virt/kvm/kvm_main.c      | 8 +++++---
>   3 files changed, 12 insertions(+), 4 deletions(-)
>
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index cd9407982366..87212d7563ae 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -3897,7 +3897,6 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>   	case KVM_CAP_ASYNC_PF_INT:
>   	case KVM_CAP_GET_TSC_KHZ:
>   	case KVM_CAP_KVMCLOCK_CTRL:
> -	case KVM_CAP_READONLY_MEM:
>   	case KVM_CAP_HYPERV_TIME:
>   	case KVM_CAP_IOAPIC_POLARITY_IGNORED:
>   	case KVM_CAP_TSC_DEADLINE_TIMER:
> @@ -4009,6 +4008,9 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
>   		if (static_call(kvm_x86_is_vm_type_supported)(KVM_X86_TDX_VM))
>   			r |= BIT(KVM_X86_TDX_VM);
>   		break;
> +	case KVM_CAP_READONLY_MEM:
> +		r = kvm && kvm->readonly_mem_unsupported ? 0 : 1;
> +		break;
>   	default:
>   		break;
>   	}
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index ddd4d0f68cdf..7ee7104b4b59 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -597,6 +597,10 @@ struct kvm {
>   	unsigned int max_halt_poll_ns;
>   	u32 dirty_ring_size;
>   
> +#ifdef __KVM_HAVE_READONLY_MEM
> +	bool readonly_mem_unsupported;
> +#endif
> +
>   	bool vm_bugged;
>   };
>   
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 52d40ea75749..63d0c2833913 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -1258,12 +1258,14 @@ static void update_memslots(struct kvm_memslots *slots,
>   	}
>   }
>   
> -static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
> +static int check_memory_region_flags(struct kvm *kvm,
> +				     const struct kvm_userspace_memory_region *mem)
>   {
>   	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
>   
>   #ifdef __KVM_HAVE_READONLY_MEM
> -	valid_flags |= KVM_MEM_READONLY;
> +	if (!kvm->readonly_mem_unsupported)
> +		valid_flags |= KVM_MEM_READONLY;
>   #endif
>   
>   	if (mem->flags & ~valid_flags)
> @@ -1436,7 +1438,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
>   	int as_id, id;
>   	int r;
>   
> -	r = check_memory_region_flags(mem);
> +	r = check_memory_region_flags(kvm, mem);
>   	if (r)
>   		return r;
>   

For all these flags, which of these limitations will be common to SEV-ES 
and SEV-SNP (ExtINT injection, MCE injection, changing TSC, read-only 
memory, dirty logging)?  Would it make sense to use vm_type instead of 
all of them?  I guess this also guides the choice of whether to use a 
single vm-type for TDX and SEV-SNP or two.  Probably two is better, and 
there can be static inline bool functions to derive the support flags 
from the vm-type.

Paolo