Re: [PATCH v8 11/11] x86/tdx: Handle CPUID via #VE

[Josh Poimboeuf]

On Mon, Oct 04, 2021 at 07:52:05PM -0700, Kuppuswamy Sathyanarayanan wrote:
> +static u64 tdx_handle_cpuid(struct pt_regs *regs)
> +{
> +	struct tdx_hypercall_output out = {0};
> +	u64 ret;
> +
> +	/*
> +	 * Emulate CPUID instruction via hypercall. More info about
> +	 * ABI can be found in TDX Guest-Host-Communication Interface
> +	 * (GHCI), section titled "VP.VMCALL<Instruction.CPUID>".
> +	 */
> +	ret = _tdx_hypercall(EXIT_REASON_CPUID, regs->ax, regs->cx, 0, 0, &out);
> +
> +	/*
> +	 * As per TDX GHCI CPUID ABI, r12-r15 registers contains contents of
> +	 * EAX, EBX, ECX, EDX registers after CPUID instruction execution.
> +	 * So copy the register contents back to pt_regs.
> +	 */
> +	regs->ax = out.r12;
> +	regs->bx = out.r13;
> +	regs->cx = out.r14;
> +	regs->dx = out.r15;

Does it still make sense to save the regs if _tdx_hypercall() returns an
error?

> +
> +	return ret;

Also I'm wondering about error handling for all these _tdx_hypercall()
wrapper functions which are called by the #VE handler.

First, there are some inconsistencies in whether and how they return the
r10 error.

- _tdx_halt() warns and doesn't return anything.

- tdx_read_msr_safe() and tdx_write_msr_safe() convert all errors to -EIO.

- tdx_handle_cpuid() returns the raw vmcall error.

Second, as far as I can tell, the #VE handler doesn't check the actual
return code value, other than checking for non-zero.  Should it at least
be printed in a warning?

-- 
Josh