Re: [PATCH v3 00/16] TDX: Enable Dynamic PAMT

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

On Mon, 2025-09-29 at 09:22 -0700, Dave Hansen wrote:
> On 9/29/25 04:17, Kiryl Shutsemau wrote:
> > > Do you have any suggestions for a better approach?
> > > 
> > > e.g., could the PAMT pages be allocated from a dedicated pool that ensures they
> > > reside in different 2MB ranges from guest private pages and TD control pages?

I guess the TDX module can get away with doing them in whatever random order
because they are try-locks. Perhaps we could take them in PFN order? If need be
the TDX module could do the same.

> > It can work: allocate 2M a time for PAMT and piecemeal it to TDX module
> > as needed. But it means if 2M allocation is failed, TDX is not functional.
> > 
> > Maybe just use a dedicated kmem_cache for PAMT allocations. Although, I
> > am not sure if there's a way to specify to kmem_cache what pages to ask
> > from page allocator.
> 
> That seems a bit obtuse rather than just respecting normal lock ordering
> rules. No?

It might serve a purpose of proving that scalability is possible. I was thinking
if we had line of sight to improving it we could go with the "simple" solution
and wait until there is a problem. Is it reasonable?

But locking issues and state duplication between the TDX module and kernel are a
recurring pattern. The similar KVM MMU issue was probably the most contentious
design issue we had in the TDX base enabling together with the TD configuration
API. To me DPAMT is just more proof that this is not a rare conflict, but a
fundamental problem. If we do need to find a way to improve the DPAMT
scalability, I want to say we should actually think about the generic case and
try to solve that. But that needs much more thought...