Re: [PATCH 4/4] KVM: selftests: Check VMPTRLD with active eVMCS

[Sean Christopherson <seanjc@google.com>]

On Mon, Aug 17, 2026, Vitaly Kuznetsov wrote:
> sashiko-bot@kernel.org writes:
> 
> > Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> > - [Medium] The hardcoded `VMPTRLD_INSN_SIZE` of 5 bytes is brittle
> > because `vmptrld` is executed using an inline assembly memory operand
> > constraint (`"m"`), which results in a variable-sized instruction
> > depending on compiler optimizations and register allocation.
> 
> True that, but afair we don't currently have an instruction decoder in
> selftests. I see three ways to move forward:
> 
> - Calculate the size of the instruction with two labels, something like
>   (untested):
> 
>     static inline int __vmptrld(u64 vmcs_pa, int *insn_size)
>     {
>             u8 ret;
> 
>             __asm__ __volatile__ ("movl $(2f-1f), %[size];"
>                                   "1: vmptrld %[pa]; 2:"
>                                   "setna %[ret]"
>                     : [ret]"=rm"(ret), [size]"=m"(*insn_size)
>                     : [pa]"m"(vmcs_pa)
>                     : "cc", "memory");
> 
>             return ret;
>     }
> 
>   and use insn_size in the tests.
> 
> - Stuf __vmptrld() with NOPs after vmptrld and keep the constant jump,
>   just make it loner (e.g. 10 bytes should realistically be enough).
> 
> - Make peace with the fragility (it's just a selftest after all) and
>   leave a comment in vmptrld() that changing the asm there make break
>   the assumption.

Option D: use KVM_ASM_SAFE() to do the heavy lifting.  Completely untested, but
something like so:

static inline int vmptrst_safe(u64 *value)
{
	u64 error_code;
	u8 vector;
	int ret;

	asm volatile(KVM_ASM_SAFE("vmptrst %[value]")
		     "\n\tsetna %[ret]"
		     : KVM_ASM_SAFE_OUTPUTS(vector, error_code),
		       [value]"=m"(*value), [ret]"=rm"(ret)
		     :
		     : "cc", "memory");

	return vector ? vector : ret ? -EINVAL : 0;
}

> Any preferences? Personally, I'm feeling adventurous and can go with the
> first one - calculating the exact size with labels.