Re: [PATCH v3 24/25] KVM: TDX: Guard VM state transitions with "all" the locks

["Huang, Kai" <kai.huang@intel.com>]

On Mon, 2025-10-27 at 17:37 -0700, Sean Christopherson wrote:
> On Tue, Oct 28, 2025, Kai Huang wrote:
> > On Thu, 2025-10-16 at 17:32 -0700, Sean Christopherson wrote:
> > > @@ -2781,8 +2827,6 @@ int tdx_vm_ioctl(struct kvm *kvm, void __user *argp)
> > >  	if (r)
> > >  		return r;
> > >  
> > > -	guard(mutex)(&kvm->lock);
> > > -
> > >  	switch (tdx_cmd.id) {
> > >  	case KVM_TDX_CAPABILITIES:
> > >  		r = tdx_get_capabilities(&tdx_cmd);
> > 
> > IIRC, this patch removes grabbing the kvm->lock in tdx_vm_ioctl() but only
> > adds the "big hammer" to tdx_td_init() and tdx_td_finalize(), so the
> > tdx_get_capabilities() lost holding the kvm->lock.
> 
> Ooh, yeah, nice catch, that is indeed silly and unnecessary churn.
> 
> > As replied earlier, I think we can just hold the "big hammer" in
> > tdx_vm_ioctl()?
> 
> Actually, I think we can have our cake and eat it too.  With this slotted in as
> a prep patch, the big hammer can land directly in tdx_vm_ioctl(), without any
> change in functionality for KVM_TDX_CAPABILITIES.
> 
> --
> From: Sean Christopherson <seanjc@google.com>
> Date: Mon, 27 Oct 2025 17:32:34 -0700
> Subject: [PATCH] KVM: TDX: Don't copy "cmd" back to userspace for
>  KVM_TDX_CAPABILITIES
> 
> Don't copy the kvm_tdx_cmd structure back to userspace when handling
> KVM_TDX_CAPABILITIES, as tdx_get_capabilities() doesn't modify hw_error or
> any other fields.
> 
> Opportunistically hoist the call to tdx_get_capabilities() outside of the
> kvm->lock critical section, as getting the capabilities doesn't touch the
> VM in any way, e.g. doesn't even take @kvm.
> 
> Suggested-by: Kai Huang <kai.huang@intel.com>
> Signed-off-by: Sean Christopherson <seanjc@google.com>
> ---
>  arch/x86/kvm/vmx/tdx.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kvm/vmx/tdx.c b/arch/x86/kvm/vmx/tdx.c
> index 1642da9c1fa9..43c0c3f6a8c0 100644
> --- a/arch/x86/kvm/vmx/tdx.c
> +++ b/arch/x86/kvm/vmx/tdx.c
> @@ -2807,12 +2807,12 @@ int tdx_vm_ioctl(struct kvm *kvm, void __user *argp)
>  	if (r)
>  		return r;
>  
> +	if (tdx_cmd.id == KVM_TDX_CAPABILITIES)
> +		return tdx_get_capabilities(&tdx_cmd);
> +

OK fine to me. :-)

Maybe add a comment saying tdx_get_capabilities() doesn't copy any data 
back to kvm_tdx_cmd structure, and doesn't need to take @kvm?  This gives
people some notice if they want to change in the future (not sure whether
any change will happen though).

>  	guard(mutex)(&kvm->lock);
>  
>  	switch (tdx_cmd.id) {
> -	case KVM_TDX_CAPABILITIES:
> -		r = tdx_get_capabilities(&tdx_cmd);
> -		break;
>  	case KVM_TDX_INIT_VM:
>  		r = tdx_td_init(kvm, &tdx_cmd);
>  		break;
> 
> base-commit: 672537233b8da2c29dca7154bf3a3211af7f6128
> --