Re: [PATCH v5 07/22] x86/virt/tdx: Implement SEAMCALL function

From: Kai Huang
Date: Tue Jul 26 2022 - 20:35:02 EST


On Thu, 2022-07-21 at 13:52 +1200, Kai Huang wrote:
> > > BUT, since as the first step, we cannot get the CMR during kernel boot (as
> > > it
> > > requires additional code to put CPU into VMX operation), I think for now
> > > we can
> > > handle ACPI memory hotplug in below way:
> > >
> > > - For memory hot-removal, we do nothing.
> >
> > This doesn't seem right to me.  *If* we get a known-bogus hot-remove
> > event, we need to reject it.  Remember, removal is a two-step process.
>
> If so, we need to reject the (CMR) memory offline.  Or we just BUG() in the
> ACPI
> memory removal  callback?
>
> But either way this will requires us to get the CMRs during kernel boot.
>
> Do you think we need to add this support in the first series?

Hi Dave,

In terms of whether we should get CMRs during kernel boot (which requires we do
VMXON/VMXOFF during kernel boot around SEAMCALL), I forgot one thing:

Technically, ACPI memory hotplug is related to whether TDX is enabled in BIOS,
but not related to whether TDX module is loaded or not. With doing
VMXON/VMXOFF, we can get CMRs during kernel boot by calling P-SEAMLDR's
SEAMCALL. But theoretically, from TDX architecture's point of view, the P-
SEAMLDR may not be loaded even TDX is enabled by BIOS (in practice, the P-
SEAMLDR is always loaded by BIOS when TDX is enabled), in which case there's no
way we can get CMRs. But in this case, I think we can just treat TDX isn't
enabled by BIOS as kernel should never try to load P-SEAMLDR.

Other advantages of being able to do VMXON/VMXOFF and getting CMRs during kernel
boot:

1) We can just shut down the TDX module in kexec();
2) We can choose to trim any non-CMR memory out of memblock.memory instead of
having to manually verify all memory regions in memblock are CMR memory.

Comments?