Re: [PATCH] x86/virt/tdx: Add ENDBR for low level SEAMCALL assembly functions

["Huang, Kai" <kai.huang@intel.com>]

On Thu, 2025-06-05 at 16:59 +0200, Peter Zijlstra wrote:
> On Thu, Jun 05, 2025 at 07:54:48AM -0700, Dave Hansen wrote:
> > On 6/5/25 07:21, Peter Zijlstra wrote:
> > > -static inline u64 sc_retry(sc_func_t func, u64 fn,
> > > -			   struct tdx_module_args *args)
> > > +static __always_inline u64 sc_retry(const sc_func_t func, u64 fn,
> > > +				    struct tdx_module_args *args)
> > >  {
> > >  	int retry = RDRAND_RETRY_LOOPS;
> > 
> > Practically, I can see how this works. If the compiler doesn't inline
> > sc_retry(), it stops being able to guarantee that the function pointer
> > value is known.
> > 
> > My only worry is that the compiler decides to be stupid for some other
> > reason, it could start generating indirect calls again.
> > 
> > Are we confident that __always_inline will keep the compiler from
> > generating indirect calls?
> 
> Fairly sure. I've used this elsewhere too.
> 
> Also, if it ever decides to be that stupid, I'm going to log a bug and
> call it broken.

Thanks Peter for helping.

I verified Peter's change, and I confirm that fixes those warnings.  Both
gcc and clang (with CFI) build successfully w/o those warnings.

Hi Dave,

Just wondering are you happy with Peter's change?