Re: [PATCH v14 09/21] KVM: guest_memfd: Track guest_memfd mmap support in memslot

[David Hildenbrand <david@redhat.com>]

On 16.07.25 10:21, Fuad Tabba wrote:
> Hi Xiaoyao,
> 
> On Wed, 16 Jul 2025 at 07:11, Xiaoyao Li <xiaoyao.li@intel.com> wrote:
>>
>> On 7/15/2025 5:33 PM, Fuad Tabba wrote:
>>> Add a new internal flag, KVM_MEMSLOT_GMEM_ONLY, to the top half of
>>> memslot->flags. This flag tracks when a guest_memfd-backed memory slot
>>> supports host userspace mmap operations. It's strictly for KVM's
>>> internal use.
>>
>> I would expect some clarification of why naming it with
>> KVM_MEMSLOT_GMEM_ONLY, not something like KVM_MEMSLOT_GMEM_MMAP_ENABLED
>>
>> There was a patch to check the userspace_addr of the memslot refers to
>> the same memory as guest memfd[1], but that patch was dropped. Without
>> the background that when guest memfd is mmapable, userspace doesn't need
>> to provide separate memory via userspace_addr, it's hard to understand
>> and accept the name of GMEM_ONLY.
> 
> The commit message could have clarified this a bit more. Regarding the
> rationale for the naming, there have been various threads and live
> discussions in the biweekly guest_memfd meeting . Instead of rehashing
> the discussion here, I can refer you to a couple [1, 2].

Maybe clarify here:

"Add a new internal flag, KVM_MEMSLOT_GMEM_ONLY, to the top half of
memslot->flags. This flag tracks when a guest_memfd-backed memory slot
supports host userspace mmap operations, which implies that all memory, 
not just private memory for CoCo VMs, is consumed through guest_memfd: 
"gmem only"

And add a pointer to the list discussion.

-- 
Cheers,

David / dhildenb