RE: [PATCH 5/6] x86/hyperv: Support hypercalls for TDX guests

From: Dexuan Cui
Date: Mon Nov 28 2022 - 14:03:47 EST


> From: Dave Hansen <dave.hansen@xxxxxxxxx>
> Sent: Monday, November 28, 2022 7:22 AM
> [...]
> On 11/27/22 16:58, Dexuan Cui wrote:
> > +u64 hv_tdx_hypercall(u64 control, u64 input_addr, u64 output_addr)
> > +{
> > + struct tdx_hypercall_args args = { };
> > +
> > + if (!(control & HV_HYPERCALL_FAST_BIT)) {
> > + if (input_addr)
> > + input_addr += ms_hyperv.shared_gpa_boundary;
> > +
> > + if (output_addr)
> > + output_addr += ms_hyperv.shared_gpa_boundary;
> > + }
>
> This:
> [...]
> makes it sound like HV_HYPERCALL_FAST_BIT says whether arguments go in
> registers (fast) or memory (slow). But this hv_tdx_hypercall() function
> looks like it takes addresses only.

Good point! When hv_tdx_hypercall() is called from hv_do_fast_hypercall8()
and hv_do_fast_hypercall16(), actually the two parameters are not memory
addresses. I'll rename the parameters to param1 and param2.

I also realized I need to export the function for drivers.

> *Is* there a register based calling convention to make Hyper-V
> hypercalls when running under TDX?

When hv_tdx_hypercall() is called from hv_do_fast_hypercall8()
and hv_do_fast_hypercall16(), the params are indeed passed via registers
rather than memory.

> Also, is this the output address manipulation fundamentally different from:
>
> output_addr = cc_mkdec(output_addr);
>
> ? Decrypted addresses are the shared ones, right?
Yes.

Good point -- I'll use the updated version:

u64 hv_tdx_hypercall(u64 control, u64 param1, u64 param2)
{
struct tdx_hypercall_args args = { };

if (!(control & HV_HYPERCALL_FAST_BIT)) {
if (param1)
param1 = cc_mkdec(param1);

if (param2)
param2 = cc_mkdec(param2);
}

args.r10 = control;
args.rdx = param1;
args.r8 = param2;

(void)__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT);

return args.r11;
}
EXPORT_SYMBOL_GPL(hv_tdx_hypercall);