Re: [PATCH v10 21/45] x86/mm: Add support to validate memory when changing C-bit

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 2/13/22 06:15, Borislav Petkov wrote:
On Fri, Feb 11, 2022 at 11:27:54AM -0600, Brijesh Singh wrote:
Simply have them always present. They will have !0 values on the
respective guest types and 0 otherwise. This should simplify a lot of
code and another unconditionally present u64 won't be the end of the
world.

Any other aspect I'm missing?

I think that's mostly about it. IIUC, the recommendation is to define a
new callback in x86_platform_op. The callback will be invoked
unconditionally; The default implementation for this callback is NOP;
The TDX and SEV will override with the platform specific implementation.
I think we may able to handle everything in one callback hook but having
pre and post will be a more desirable. Here is why I am thinking so:

* On SNP, the page must be invalidated before clearing the _PAGE_ENC
from the page table attribute

* On SNP, the page must be validated after setting the _PAGE_ENC in the
page table attribute.

Right, we could have a pre- and post- callback, if that would make
things simpler/clearer.

Also, in thinking further about the encryption mask, we could make it a
*single*, *global* variable called cc_mask which each guest type sets it
as it wants to.

Then, it would use it in the vendor-specific encrypt/decrypt helpers
accordingly and that would simplify a lot of code. And we can get rid of
all the ifdeffery around it too.

So I think the way to go should be we do the common functionality, I
queue it on the common tip:x86/cc branch and then SNP and TDX will be
both based ontop of it.

Thoughts?

I think there were a lot of assumptions that only SME/SEV would set sme_me_mask and that is used, for example, in the cc_platform_has() routine to figure out whether we're AMD or Intel. If you go the cc_mask route, I think we'll need to add a cc_vendor variable that would then be checked in cc_platform_has(). All other uses of sme_me_mask would need to be audited to see whether cc_vendor would need to be checked, too.

Thanks,
Tom





[Index of Archives]     [Linux Kernel Development]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux