Re: [PATCH 1/2] x86/virt/tdx: Use PFN directly for mapping guest private memory

["Edgecombe, Rick P" <rick.p.edgecombe@intel.com>]

On Fri, 2026-03-20 at 10:38 -0700, Dave Hansen wrote:
> Can we add something that will BUG_ON() or fail to compile when the huge
> page support comes around?
> 
> I'd much rather have:
> 
> 	BUG_ON(level != PG_LEVEL_4K);
> 	tdx_clflush_pfn(pfn);
> 
> That go implementing a level argument to tdx_clflush_pfn() now. Then
> nobody has to remember. The list to remember is in the code.

I like it, but I'm afraid to add BUG_ON()s. Could we do a WARN instead?

Especially since... the ridiculous thing about this is that the clflush is only
needed if CLFLUSH_BEFORE_ALLOC is set in the tdx metadata, which we have yet to
see in any modules. The spec is apparently to give TDX module flexibility for
the future.

In the base series we debated just ignoring it, but at the time it was simpler
to just always flush. So if CLFLUSH_BEFORE_ALLOC never comes along, it is
possible the function will never do any useful work in its life. Tough case for
a BUG_ON().