mirror of
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
synced 2026-09-12 01:25:43 -04:00
2abcdf03fd
When calling into arch code to make the underlying memory private, i.e. to
assign memory to the VM in SNP's RMP table, assign/convert *exactly* the
range of memory that can be mapped into the guest for the current page
fault, instead of aggressively converting/assigning the entire folio. For
SNP, the mapping size in the stage-2 page tables (Nested Page Tables, NPT)
must be at least the size of the corresponding RMP entry, e.g. assigning a
2MiB mapping in the RMP when it can only be mapped at 4KiB granualarity
will ultimate result in another page fault (#NPF for SNP) to "smash" the
RMP down to the correct mapping size.
Assigning the entire folio was necessary back when guest_memfd tracked
preparedness, which was done on a per-folio basis. At the time, it made
sense to do per-folio tracking/preparation, because tracking per-folio
meant guest_memfd didn't need to add a separate data structure to track
that information, and doing per-folio tracking only works if the entire
folio is prepared (or not).
Now that guest_memfd no longer does preparation tracking (see commit
8622ef0570 ("KVM: guest_memfd: Remove preparation tracking")), in favor
having SNP query the RMP, per-folio preparation, i.e. per-folio conversions
to private, doesn't make any sense.
*If* SNP allowed the RMP size to be greater than the NPT size, then
per-folio conversion could theoretically provide marginal value, as it
would allow KVM to assign a hugepage in the RMP even if it can only be
mapped into the NPT with a smaller page, e.g. because of memslot alignment.
The documentation of that reasoning would be something like this:
/*
* If the memory is private from KVM's perspective, and hardware tracks
* VM-assigned private memory in a dedicated data structure, i.e. not
* in the stage-2 page tables, then call into arch code to assign the
* entire folio to the guest. Assigning the entire folio, e.g. instead
* of only the memory being mapped into the guest, allows KVM to assign
* an entire hugepage of memory in the out-of-band structure even if
* KVM can only map a smaller page size into the MMU, e.g. because the
* gmem hugepage is spread across multiple memslots.
*/
But even *if* a future SNP implementation supported that behavior, the
value added would be dubious, as having a huge folio that is fully private,
but can only be mapped at a smaller granularity, would be rare. E.g. maybe
for memory at the top of lower DRAM that has holes for non-RAM assets?
So, convert/assign exactly what guest_memfd allows the caller to map to
simplify the guest_memfd code and provide a (super) minor performance
optimization for SNP. E.g. once hugepage support comes along, guest_memfd
will only need a single flow to compute "how much memory can be assigned
and at what size".
Reviewed-by: Ackerley Tng <ackerleytng@google.com>
Link: https://patch.msgid.link/20260723210811.72720-10-seanjc@google.com
Signed-off-by: Sean Christopherson <seanjc@google.com>