From: Sathyanarayanan Kuppuswamy sathyanarayanan.kuppuswamy@linux.intel.com Sent: Tuesday, May 30, 2023 6:22 AM
Hi,
On 5/30/23 5:57 AM, Tom Lendacky wrote:
On 5/29/23 19:57, Kirill A. Shutemov wrote:
On Fri, May 26, 2023 at 03:10:56PM -0700, Sathyanarayanan Kuppuswamy wrote:
On 5/26/23 5:02 AM, Kirill A. Shutemov wrote:
Touching privately mapped GPA that is not properly converted to private with MapGPA and accepted leads to unrecoverable exit to VMM.
load_unaligned_zeropad() can touch memory that is not owned by the caller, but just happened to next after the owned memory.
/s/to/to be ?
Yep, my bad.
This load_unaligned_zeropad() behaviour makes it important when kernel asks VMM to convert a GPA from shared to private or back. Kernel must never have a page mapped into direct mapping (and aliases) as private when the GPA is already converted to shared or when GPA is not yet converted to private.
I am wondering whether this issue exist in the AMD code?
IMO, you can add some info on the window in set_memory_encrypted() where this race exists.
I don't think AMD affected by load_unaligned_zeropad() the same way as Intel does. But I'm not sure.
Tom, do you have any comments?
Right, shouldn't be an issue for SNP.
Thanks for confirming.
Tom -- For my education, could you elaborate on why this problem can't occur in an SEV-SNP guest? There's still a window where the direct map PTE and the RMP as maintained by the hypervisor are out-of-sync. If load_unaligned_zeropad() does a read using the direct map PTE during this out-of-sync window, isn't that going to trap to the hypervisor? How is the scenario is handled from there to provide the zeros to load_unaligned_zeropad()? I need to make sure Hyper-V is doing whatever is needed. :-)
Thanks,
Michael