Re: [PATCH v4 05/15] x86/sev: Use kernel provided SVSM Calling Areas

[Tom Lendacky]

On 5/6/24 05:09, Borislav Petkov wrote:
> Ok,
>
> I think this is very readable and clear what's going on:

I'll test it out.

> static __always_inline void issue_svsm_call(struct svsm_call *call, u8 *pending)
> {
>          register unsigned long rax asm("rax") = call->rax;
>          register unsigned long rcx asm("rcx") = call->rcx;
>          register unsigned long rdx asm("rdx") = call->rdx;
>          register unsigned long r8  asm("r8")  = call->r8;
>          register unsigned long r9  asm("r9")  = call->r9;
>
>          call->caa->call_pending = 1;
>
>          asm volatile("rep; vmmcall\n\t"
>                       : "+r" (rax), "+r" (rcx), "+r" (rdx), "+r" (r8), "+r" (r9));
>
>          xchg(pending, 1);

This isn't quite right. The xchg has to occur between pending and 
call->caa->call_pending.

Thanks,
Tom

>          call->rax_out = rax;
>          call->rcx_out = rcx;
>          call->rdx_out = rdx;
>          call->r8_out = r8;
>          call->r9_out = r9;
> }
>
> and the asm looks ok but the devil's in the detail.