Re: [PATCHv2] x86/mm: Fix memory encryption features advertisement

[Huang, Kai]

On Tue, 2024-01-16 at 13:58 +0300, kirill.shutemov@xxxxxxxxxxxxxxx wrote:
> On Tue, Jan 16, 2024 at 10:36:10AM +0000, Huang, Kai wrote:
> > On Thu, 2024-01-11 at 14:12 +0300, Kirill A. Shutemov wrote:
> > > When memory encryption is enabled, the kernel prints the encryption
> > > flavor that the system supports.
> > > 
> > > The check assumes that everything is AMD SME/SEV if it doesn't have
> > > the TDX CPU feature set.
> > > 
> > > Hyper-V vTOM sets cc_vendor to CC_VENDOR_INTEL when it runs as L2 guest
> > > on top of TDX, but not X86_FEATURE_TDX_GUEST. Hyper-V only needs memory
> > > encryption enabled for I/O without the rest of CoCo enabling.
> > > 
> > > To avoid confusion, check the cc_vendor directly.
> > > 
> > > Possible alternative is to completely removing the print statement.
> > > For a regular TDX guest, the kernel already prints a message indicating
> > > that it is booting on TDX. Similarly, AMD and Hyper-V can also display
> > > a message during their enumeration process.
> > > 
> > > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > > Cc: Dexuan Cui <decui@xxxxxxxxxxxxx>
> > > Cc: Jeremi Piotrowski <jpiotrowski@xxxxxxxxxxxxxxxxxxx>
> > 
> > Seems this fix is for userspace running in hyperv environment being able to use
> > some easy grep to get which coco vendor it is running on?
> 
> Making decision in userspace by	grepping dmesg is bad idea and nobody
> should do this. It can easily give false result: dmesg is not ABI, format
> can change and ring buffer has finite size, the message can be overridden.
> 
> If we need a way for userspace to discover which CoCo environment it runs
> on, we need proper ABI for that. Maybe sysfs file or something.

Yeah agreed.  :-)