Re: [PATCH v1 2/3] kvm: svm: Add kvm_fast_pio_in support

[Paolo Bonzini <pbonzini@redhat.com>]

----- Original Message -----
> From: "Tom Lendacky" <thomas.lendacky@amd.com>
> To: "Paolo Bonzini" <pbonzini@redhat.com>, "Brijesh Singh" <brijesh.singh@amd.com>, kvm@vger.kernel.org
> Cc: rkrcmar@redhat.com, joro@8bytes.org, x86@kernel.org, linux-kernel@vger.kernel.org, mingo@redhat.com,
> hpa@zytor.com, tglx@linutronix.de, bp@suse.de
> Sent: Tuesday, November 22, 2016 3:25:52 PM
> Subject: Re: [PATCH v1 2/3] kvm: svm: Add kvm_fast_pio_in support
> 
> On 11/21/2016 8:50 AM, Paolo Bonzini wrote:
> > 
> > 
> > On 14/11/2016 23:15, Brijesh Singh wrote:
> >> +	/* For size less than 4 we merge, else we zero extend */
> >> +	val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
> > 
> > Are you sure it shouldn't always zero extend the high 32-bits?  So "val"
> > should be declared as u32.
> > 
> > Paolo
> 
> It should only zero extend when dealing with a 32-bit operation. Any use
> of 8 or 16 bit registers leaves the upper 56 or 48 bits as is (see
> section 3.1.2.3 in http://support.amd.com/TechDocs/24592.pdf).

Duh, right, see also assign_register in arch/x86/kvm/emulate.c.

Paolo

> Thanks,
> Tom
> 
> > 
> >> +	ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
> >> +				       &val, 1);
> >> +	if (ret) {
> >> +		kvm_register_write(vcpu, VCPU_REGS_RAX, val);
> >> +		return ret;
> >> +	}
>